Peregrine Connects Existing City Data to Power Public Safety Software

Peregrine连接城市既有数据，赋能公共安全软件 在不增设硬件传感器的情况下打通市政数据，使公共安全机构能够在保护隐私和本地数据主权的同时开展AI分析。 Peregrine创始人Nick Noone和Ben Rudolph阐释了该平台如何通过聚合城市现有的零散数据而非部署新的监控硬件，来颠覆传统的公共安全科技模式。结合他们在Palantir从事高风险情报行动以及在联合国参与国际难民救济的过往背景，两人详细介绍了其前沿部署工程方法如何与地方机构建立深厚信任。他们梳理了AI与自主智能体的实际业务应用，包括协助核实复杂平反案件的悬案智能体、针对仇恨言论威胁的多系统模式分析以及灾害响应建模。在此过程中，他们始终强调数据主权、伦理保障以及降低软件成本，从而让小型社区同样能够使用先进技术。 Peregrine 彻底颠覆了依赖硬件传感器收集新数据的传统公共安全模式，通过聚合与治理城市既有的孤立数据资产，在严守客户数据主权的同时杜绝构建中心化监控网络。 (18:16) 在遭遇二十多次拒绝后，创始人以极其谦逊的态度和一线实地跟班换取了加州圣巴勃罗警察局（San Pablo PD）的准入机会，并将前线部署工程师（FDE）视为产品研发与原型设计的核心引擎。 (10:45) 平台推动公共安全机构从基础搜索迈向深度多系统研判，曾通过合成调度记录与历史天气数据查明佛州激增的上百起水上救援成因，并高效识别针对犹太教堂的非结构化仇恨犯罪线索。 (22:57) 目前 Peregrine 约 90% 的数据集成由自主智能体完成；其冷案智能体（Cold Case Agent）不仅独立复现了耗时数月的冤案平反结论，还在威斯康星州包含 300GB 复杂证据的凶案中精准定位出关键基站通信记录。 (29:59) 公司奉行“反网络效应”理念并坚决反对数据池化，致力于打破硅谷传统的八位数高昂合同壁垒，以大幅降低的边际成本为中小型及资源匮乏的基层公共安全部门提供全栈式数据基础设施。 (40:00)

Peregrine founders Nick Noone and Ben Rudolph argue that public safety and civic health depend on solving government data fragmentation through integration and strict local sovereignty, rather than expanding surveillance hardware or centralized data pooling. Forward Deployment and Humility in Public Institutions (0:00) Peregrine was built on the premise that municipal institutions fail not from lack of effort, but from disconnected data architectures. Drawing on Noone’s experience leading Palantir’s SOCOM unit and Rudolph’s field work with UNHCR and Dimagi, the founders adopted a forward-deployed engineering model. This approach requires engineers to embed directly inside agencies, suppress personal ego, and co-own problems alongside career public servants. In 2018, following dozens of rejections, the founders secured badges and desk space with Commander Brian Bubar at the San Pablo Police Department by taking a collaborative, exploratory stance rather than pitching a finished product. Operating through the political tensions of 2020 and the 'Defund the Police' movement, the founders navigated intense recruiting hurdles to build apolitical operational tools. To understand field realities, they conducted overnight ride-alongs with the Oakland Police Department and worked alongside homicide detectives, eventually broadening Peregrine's software footprint from law enforcement to fire services, EMS, health departments, and emergency management agencies. Inverting the Data Collection Business Model (18:16) Unlike legacy public safety tech providers such as Axon or Flock Safety, which generate revenue by deploying hardware sensors, automated license plate readers, and body cameras to collect proprietary data assets, Peregrine deliberately deploys zero hardware. Instead, the platform sits on top of a municipality’s pre-existing, siloed databases to aggregate and govern information the agency already owns. Positioning itself as an 'anti-network-effects' company, Peregrine rejects pooling customer data into a centralized repository. Rudolph and Noone stress that agency data belongs exclusively to that specific agency and its community. The platform relies on strict permissions and auditable boundaries to avoid surveillance overreach, replacing insecure workarounds like transporting physical hard drives while rejecting unilateral corporate mandates on controversial capabilities like facial recognition. AI-Driven Synthesis and Autonomous Agents (22:57) Peregrine expands agency capabilities beyond simple record retrieval to multi-system reasoning. In one Florida county, the platform synthesized incident logs, 911 calls, and weather data to reveal that a three-day storm system had altered coastal sand channels, exposing the physical mechanism behind a surge of over 100 water rescues. In another jurisdiction, semantic analysis connected disparate threat reports across systems to identify coordinated anti-Semitic threats targeting area synagogues. Noone notes that 95% of engineering complexity lies in pre-processing, stitching, and structuring data prior to querying to ensure factual citations. To manage vast data volumes, Peregrine allocates roughly 50% of its engineering resources to its core data platform, using autonomous orchestrator and sub-agents to author approximately 90% of its data integrations under human oversight. Operationally, specialized long-horizon agents process massive case files. In one benchmark, an agent ingested 200 to 300 gigabytes of unstructured records and independently reproduced a departmental exoneration of a wrongfully convicted individual in under an hour. In a Wisconsin homicide investigation, an agent isolated a decisive cell tower record across millions of data points to place a suspect at both the crime scene and body disposal site. Scaling Enterprise Tech to Underserved Communities (40:00) Traditional enterprise software vendors have focused on eight-figure contracts with large agencies, leaving smaller municipal departments economically unviable. Peregrine structured a vertically integrated stack—incorporating network security, permissioning, ETL pipelines, ontologies, and application interfaces—to deliver enterprise integration at price points an order of magnitude lower than legacy competitors. Forward-deployed engineers serve as the company's primary R&D engine. When field engineers encounter functionality gaps, their localized workarounds frequently become core platform features, such as parsing user comments to trigger database updates. Field strategists also rapidly generate custom applications, including hurricane simulators and interactive modeling tools that calculate the impact of new fire stations using 911 dispatch times and city budgets. Peregrine aims to serve as the institutional memory layer for over 10,000 cities over the next decade. The primary operational test remains whether the company can maintain low marginal deployment costs and absolute data sovereignty without compromising security or imposing centralized policies on local public safety agencies.

Peregrine 联合创始人 Nick Noone 与 Ben Rudolph 提出，现代公共安全机构的核心困境并非缺乏数据或监控硬件，而是既有存量系统的高度割裂与治理缺陷；该公司通过前线部署工程、严格的数据主权架构与自主智能体技术，致力于在不建立集中监控网络的前提下，为被传统巨头忽视的中小城市提供低成本、高精度的基础数据智能平台。 颠覆传统监控模式与坚持数据主权 (18:16) Peregrine 采取了与 Axon 或 Flock Safety 等传统公共安全技术厂商截然相反的商业路径。传统厂商通过铺设车牌识别仪、执法记录仪等新型硬件传感器来收集并囤积数据，以此锁定客户并追加销售硬件系统；Peregrine 则完全不部署任何新传感器，也不采集额外的原始数据，而是直接构建在地方政府既有的碎片化数据库之上，专注于对现有资产进行治理、聚合与精准交互。 创始人强调，针对公众对中央集权式监控系统（Panopticon）的担忧，公司确立了“反网络效应”（anti-network-effects）原则，坚决拒绝跨机构汇总客户数据的商业模式。Peregrine 严格保障地方机构的数据主权，坚持数据仅归属于对应部门及其社区，并实施细粒度的访问控制与防过度共享机制，以规范化协作取代以往执法人员驱车肉身运输原始硬盘的低安全性做法。在面部识别等敏感技术领域，Peregrine 拒绝向客户强制推销功能，而是作为审慎顾问尊重地方社区和法律的自主选择。 前线部署工程与高同理心的信任构建 (0:00) Peregrine 的技术哲学建立在创始团队过往的实战背景上：Nick Noone 曾主导 Palantir 的 SOCOM（特种作战司令部）中东情报部署，Ben Rudolph 则经历过联合国难民署（UNHCR）边境人道主义救援及 Dimagi 在印度全国推广的结核病治疗追踪项目。他们认为硅谷在进入公共领域时往往因傲慢而受阻，真正的“前线部署工程”（FDE）必须压制自我、以同理心深入真实环境，并完全将最终成果归功于一线公职人员。 在经历了 2017 年二十多次推介遭拒后，团队于 2018 年 2 月 26 日通过主动请教的方式，获得了加州 San Pablo 警察局指挥官 Brian Bubar 的信任并进入实地。在 2020 年“削减警察预算”（Defund the Police）运动、新冠疫情及剧烈政治极化的背景下，尽管面临工程师招聘困难，团队仍坚持驻扎基层重案调查一线。随着平台能力的验证，Peregrine 的服务范围逐步从最初的警务执法，横向拓展至消防救援、紧急医疗服务（EMS）、公共卫生及应急管理等多类市政部门。 自主智能体驱动的多系统深度研判 (22:57) 在应用层面，Peregrine 推动公共机构从基础的信息检索跃升至复杂的跨系统因果分析。例如，佛罗里达州某县单月遭遇超过 100 起异常激增的水上救援，系统通过综合分析报警日志、911 调度与气象记录，查明救援激增的主因是罕见的连续三天气象模式冲刷出近海动态沙槽，从而诱发了致命离岸流，为后续前置巡防提供了依据；在针对犹太教堂的威胁排查中，系统的语义推理能力也成功穿透了非结构化文本与非标准关键词。 为支撑庞大的整合需求，Peregrine 将约 50% 的工程力量倾注于数据平台，并将 95% 的技术复杂度放在查询前的数据预处理与关系缝合上。目前约 90% 的数据集成工作由长周期自主智能体（Autonomous Agents）在人类监督下独立完成。此外，其研发的冷案智能体（Cold Case Agent）能够处理 200 至 300GB 的非结构化多媒体与卷宗数据：该智能体不仅在 30 至 60 分钟内自主复现了此前人工耗费数月才完成的冤案平反调查，还在威斯康星州的一起命案中精准锁定了将嫌疑人与抛尸现场关联的关键基站呼叫记录（CDR）。 赋能基层机构与低边际成本扩张路径 (40:00) 与 Palantir 依赖八位数巨额合同的运作模式不同，Peregrine 选择聚焦被传统科技巨头视作无利可图的中小规模公共安全机构。通过构建涵盖网络准入、权限管控、ETL 流水线、自定义本体与开放 API 的垂直整合技术栈，Peregrine 将定制化企业级数据系统的交付成本降低了一个数量级，使无力承担数百万美元年费的地方县市同样能获取先进技术。 在该体系中，前线部署工程师（FDE）直接扮演了产品研发与创新的核心引擎。一线工程师面对现场实际操作痛点时开发的临时方案，能够快速反哺平台底层演进——例如早期针对字段编辑需求开发的临时解析逻辑直接催生了原生编辑功能；驻场人员还针对地方实际场景，快速构建出了原生的飓风模拟器，以及结合 911 响应时效与财政预算的新建消防站影响评估工具，实现了敏捷的功能原型验证。 Peregrine 的长期愿景是在十年内成为全球上万座城市的制度记忆层与底层数据基础设施。然而，在坚持不采集额外硬件数据、不汇聚中心化数据池的约束下，如何在服务极度分散且预算有限的中小城市时持续维持极低边际成本，并平衡复杂的本地法规与公众对公共安全技术的敏感监督，依然是该模式需要长期验证的课题。

## Background 本期内容由红杉资本（Sequoia Capital）于 2026 年 9 月 1 日在播客及网络平台发布。节目主要围绕 Peregrine 联合创始人 Nick Noone 与 Ben Rudolph 的访谈展开，介绍了他们如何颠覆传统公共安全科技依赖新增硬件与传感器收集数据的模式，转而通过连接各城市既有异构数据系统，在严格捍卫数据主权与公民隐私的前提下，赋能全美数百个城市的执法、消防与应急管理。 ## TL;DR Peregrine 的联合创始人分享了公司从早期遭遇 20 余次拒绝，到最终入驻圣巴勃罗警局并逐步拓展至 400 多个社区的创业历程。与一味打造集中监控帝国的硬件厂商不同，Peregrine 坚持“反网络效应”与数据主权哲学，不安装传感器或制造新数据，而是构建轻量治理层打通客户既有的孤岛信息。团队详细剖析了长周期 Agent 在自动化数据集成与悬案平反中的落地实践，以及如何依托前线部署工程师（FDE）文化，将曾经动辄千万美元的高级数据智能基础设施以极具韧性且经济可行的方式普及给中小城市。 ## Atmosphere & style 整场对话沉稳、真诚且充满反思精神，既有深耕高风险前线的硬核实战感，又流露着硅谷技术精英难得一见的谦逊与克制。两位创始人摒弃了空洞的科技浮夸辞藻，以富有同理心的叙事重现了在社会动荡与“削减警费”舆论撕裂期扎根一线的艰难岁月。听众能深切感受到他们面对数据权力和公共安全使命时的敬畏之心，整体氛围严谨务实而又极具社会理想主义色彩。 ## Bulletin - `18:16` 颠覆传统公共安全硬件垄断：与依赖摄像头和传感器无节制扩张数据采集的传统模式相反，Peregrine 坚持不装硬件、不产生新数据，专注于打通城市现存的异构孤岛数据并严控数据主权。 - `29:59` 长周期 Agent 自动化完成 90% 数据集成：通过构建高标准的确定性评测集，Peregrine 自研的后台 Agent 可连续运行数小时解析数据库架构与本体，自动编写了全平台近 90% 的异构数据集成代码。 - `31:45` 长程推理赋能重案与陈年悬案侦破：业务端开发的悬案调查 Agent 成功在 30 至 60 分钟内复现人类刑警历经数月得出的冤案平反结论，并曾在威斯康星州 300GB 杂乱证据中精确定位嫌疑人基站轨迹与抛尸现场。 - `40:00` 推动高阶定制平台降维普惠下沉市场：打破 Palantir 等巨头动辄千万美元级（八位数）的定制服务门槛，Peregrine 通过全栈垂直整合自研架构将交付成本降低一到两个数量级，赋能超 400 个中小城市基层治理。 ## Key takeaways - `00:00` 现代城市繁荣的底层基石是安全，这不仅需要物理层面的绝对安全，还需要心理层面的主观安全感；唯有搭建安全且主权独立的数字骨干，才能释放社区潜能。 - `02:07` 前线部署工程师（FDE）文化的核心在于心理上的极端所有权（Co-own the Problem）以及将荣耀全部归于基层客户，交付速度要比传统方式快 3 到 5 倍。 - `03:58` 硅谷面对传统公共机构最易犯的傲慢错误是试图在几天内推翻他人积淀数十年的体制逻辑，唯有放下优越感、保持极致同理心才能真正扎根一线。 - `05:23` 联合创始人 Ben Rudolph 在联合国难民署与 Dimagi 的基层经历表明，大量看似无解的现实社会困境本质上都是“下游数据孤岛问题”，理顺数据流转即可释放巨大社会价值。 - `10:00` 在遭遇超过 24 次无情拒绝后，创始团队凭借极具诚意与谦逊的请教姿态打动了圣巴勃罗警局指挥官 Brian Bubar，于 2018 年 2 月获得首个办公室门禁与数据访问权。 - `14:19` 在全美“削减警费”动荡与社会极化期间，团队拒绝非黑即白的标签预设，以务实、非政治化（apolitical）的底层基础设施直面复杂的现实治理灰色地带。 - `18:16` 彻底倒置传统商业模式：不同于 Axon 或 Flock 等依靠硬件传感器采集囤积数据的模式，Peregrine 坚持不买硬件、不引入新冗余数据，纯做现有异构数据的安全治理层。 - `20:00` 推行去中心化的“反网络效应”哲学：坚决反对建立中央集权式监控帝国（Panopticon），将数据所有权与审计控制权彻底归还给本地机构与社区本身。 - `22:57` 突破浅层语义搜索，迈向多源因果归因：在佛罗里达水上救援突增案中，智能体融合灾情、历史气象与微地形数据，深度推导出了罕见持续风暴侵蚀海底沙槽诱发离岸流的隐秘物理因果链。 - `26:30` AI 回答体验的冰山之下是 95% 的前置数据准备：高昂的工程研发专注于多源数据编排、本体关联清洗与可验证引用，确保大语言模型具备极高精确度。 - `29:59` 构建长周期集成 Agent 体系：通过总编排 Agent 派生子 Agent 解析本体并自动编写集成流水线，在确定性评测集监督下自主完成了 90% 的接入代码编写。 - `31:45` 悬案调查 Agent 具备强大的长程推理能力：能从 200–300GB 包含影音与扫描件的复杂非结构化卷宗中，精准抽取手机基站通话记录（CDR）等关键证据。 - `38:40` 在技术伦理与敏感领域拒绝越俎代庖：针对人脸识别等高争议技术，Peregrine 坚决反对科技巨头强加价值观，主张把合规边界与技术采用的决策权留给本地社区主权选择。 - `42:54` 以“黑暗洞穴”隐喻确立前线工程师的极端自治：充分信任前线独立贡献者（IC），将一线突破所诞生的粗糙解法（Hacky Prototyping）转化为平台核心技术原语与创新应用。 - `48:50` 服务 10,000 座城市的终局定位是成为中立的“机构记忆层”，以严谨的管家精神保护每一个城市的独特性，赋能而不施加自上而下的权威控制。 ## Detailed walkthrough ### `00:00` 重新思考城市基石：用数据连接而非监视 Peregrine 的创立源于一个核心命题：如何利用前沿技术赋能城市、县区乃至州政府，从而真正改变人们生活的环境。在构建“理想城市（awesome cities）”的需求金字塔底端，最不可或缺的基石便是安全——人们既需要客观上的物理安全，也需要在心理层面上切实“感到安全”。只有当这份基础的稳定性得以确立，城市繁荣的各种可能性才会被彻底激活。 在创业初期，创始团队一直在反复权衡一组看似对立却高度统一的命题：一方面，技术必须深入大型公共机构内部，彻底根除敏感复杂数据在流转过程中的滥用与管理不善；另一方面，如果能够在现代城市的基础设施层建立起坚实的“数字骨干（backbone infrastructure）”，就既能守护每一个个体的隐私与独立性，又能将整个社区紧密联结在一起。与传统公共安全科技公司一味依赖铺设摄像头、传感器来扩张“监控帝国”的做法相反，Peregrine 选择直接拒绝全面监控国家（surveillance state）的模式，走出了一条以保护公民自由为前提的 AI 赋能之路。 ```mermaid flowchart TD accTitle: Peregrine's Foundational Thesis accDescr: Diagram illustrating the bottom-up societal impact pyramid, starting from foundational safety to flourishing cities. subgraph Pyramid ["城市繁荣底座"] direction TB Safety["客观安全与主观安全感\n(基石支撑)"] Infrastructure["安全且主权独立的骨干基础设施\n(消除信息滥用 / 保护公民隐私)"] AwesomeCities["蓬勃发展的现代城市与社区\n(释放个体潜能 / 带来社会凝聚力)"] Safety --> Infrastructure --> AwesomeCities end ``` ### `02:07` 什么是“前线部署工程师”（Forward Deployed Engineering） 谈及在 Palantir 执掌特种作战司令部（SOCOM）业务单元并亲赴中东参与极高风险情报行动的经历，联合创始人 Nick Noone 坦言，在加入 Palantir 之前，他对“前线部署工程（Forward Deployed Engineering, FDE）”这个概念毫无认知。他当时只看到身边的工程师们频繁奔赴客户一线现场，投入深不可测的业务泥潭，随后带着震撼人心的实战故事归来。这种深入客户前线、全权肩负客户难题的历程，对他而言就像是“一生的冒险”。 Nick 总结出早期 FDE 经历中最核心的两条法则，这也是后来贯穿 Peregrine 文化的灵魂支柱： 1. **心理上的极端所有权（Co-own the Problem）**：团队必须从心理上彻底认领客户的难题。在 Peregrine 内部，这被称为“为客户、并代表客户一路挺进到最终成果（getting all the way to the outcome）”。技术本身只是达成目标的一部分，真正的核心是不惜一切代价拿下结果，且交付速度必须比任何个人或传统团队快上 3 到 5 倍。 2. **将荣耀归于客户（It's the Customer's Win）**：团队必须清醒认识到，最终的胜利完全属于客户，而不是科技公司自己的炫技。从心底里为客户机构以及坐在谈判桌对面的基层公仆所取得的突破而自豪，正是这种文化最具魔力的地方。 ### `03:58` 硅谷最容易犯的傲慢错误 当被问及硅谷在面对政府与公共机构时究竟做错了什么，Nick 指出，顶级名校背景（如全美 Top 5 或 Top 10 名校）的聪明工程师们最容易陷入“技术自负”的陷阱。当他们踏入像洛杉矶警察局（LAPD）这样历史沉淀极深、错综复杂的机构时，往往试图在极短的时间内“看懂”别人积淀了 30 年的体制上下文。 这种体制上下文不仅存在于数据接口层面，更深植于人性与组织架构之中： - 这个庞大的机构平日究竟如何运转？ - 权力与决策链条是如何实际流动的？ - 各个业务部门之间如何协作或制衡？ 在很多公职人员将毕生心血都奉献给某一个特定职位的机构里，硅谷那种自以为是、“企图在几天内打破一切并快速迭代”的姿态，不仅显得极其不切实际，更容易被一线人员直接解读为赤裸裸的傲慢。硅谷常常忘记，最难做到也最宝贵的品质，恰恰是“放下自己的聪明才智、放下自身的专业光环、悬置个人的优越感（suspend our ego）”，以极高的同理心和超乎寻常的耐心，真正将自己沉浸到客户的现实处境中去。 ### `05:23` 从 UNHCR 到 Dimagi：下游社会难题本质是“数据孤岛” 与 Nick 深入中东危险战场截然不同，联合创始人 Ben Rudolph 毕业后走向了另一个极端的现实世界。当时他手头握有两个截然相反的 Offer：一个是加入当时规模尚小的明星独角兽 Airbnb，另一个则是加入联合国难民署（UNHCR），奔赴战火纷飞的边境线为逃离家园的难民提供人道援助。出于“如何用技术产生真实社会影响力”的执念，Ben 毅然选择了后者。 在被派驻到苏丹边境与哥伦比亚边境期间，Ben 亲身体验了极其残酷的人道主义危机，同时也见证了技术在现实面前的无力与潜力： - **UNHCR 的数据断层**：联合国难民署虽然做着令人肃然起敬的工作，但内部机构却处于深度脱节的状态。宝贵的难民救援数据往往散落在一个个互不通气的 Excel 电子表格里，极难提取出有效洞察。这让他确立了一个核心认知：**大量看似无解的现实社会困境，追根溯源都是“下游数据问题（downstream data problems）”**，只要理顺数据，就能释放出巨大的社会价值。 - **Dimagi 与印度结核病防治**：离开 UNHCR 后，Ben 加入了一家鲜为人知但深耕下沉市场的科技公司 Dimagi，专注于为全球资源极度匮乏的地区打造最后一公里医疗数字化方案。在与印度政府合作的项目中，Ben 参与开发了帮助印度农村地区患者按时服用抗结核药物的移动应用。这项技术最终成功在印度全国范围内大面积推广应用，拯救了无数生命。 这段经历彻底坚定了 Ben 的技术价值观，也成为日后 Peregrine 核心理念的雏形：技术不能只悬浮在硅谷的象牙塔里，而应当扎根于人们生活的现实社区中，为公共医疗、应急与市政服务提供扎扎实实的支撑。 ### `08:25` 为什么聚焦城市安全？2016 年的体操队聚会 2016 年，大学毕业已近十年的 Nick 和 Ben 在一次前体操队队友的集体度假中重聚。两人随后在旧金山长谈，探讨携手创立一家公司的可能性。 当时两人的经历形成了极具张力的互补： - Nick 常年往返于巴格达与中东险境，在国家安全和美国联邦政府情报系统拥有深厚沉淀； - Ben 则奔波于非洲和全球偏远地区，在国际人道主义援助和公共卫生技术落地方面积累了丰富经验。 坐在一起审视世界时，他们发现了一个再朴素不过的事实：城市才是绝大多数普通人生活、工作、繁衍生息的地方。当他们开始拆解如何真正服务好美国本土城市时，所有思考最终都收敛到了最底层的硬需求——公共安全。如果缺乏安全感，城市的各项活力都无从谈起；而一旦解决了安全与信息流转失控的问题，一个兼顾个人自由与集体福祉的现代城市生态便能破土而出。 ### `10:00` 两次碰壁与圣巴勃罗警局的破冰契机 Peregrine 的创业初衷在于消除复杂组织在处理超高敏感度数据时的不当操作。创始团队相信，只要为现代城市构建起坚实的底层数据基础设施，就能在保护个体隐私与独特性的同时，将城市中的各方力量凝聚起来——当所有人注视着同一份真实透明的信息时，大家对美好城市与安全的诉求其实是高度一致的。 然而，将这一理念落地的过程充满荆棘。在北加州的圣巴勃罗警察局（San Pablo PD）最终接纳他们之前，团队遭到了超过两打（24 次以上）的无情拒绝。直到 2018 年 2 月 26 日——距离公司注册成立仅过去约一个月——圣巴勃罗警局才正式发给他们门禁卡，在警局内部安排了工位，并开放了数据信息的访问权限，Peregrine 的技术研发才真正破土动工。 打破僵局的转折点源于一次充满敬意的“冷启动”沟通。团队在调研城市级公共安全专家时，读到了一篇关于圣巴勃罗警局年轻指挥官布赖恩·布巴尔（Brian Bubar）的深度报道。布巴尔在这套讲求资历与年资的半军事化警务体制中脱颖而出，凭借过人的才干快速晋升。他在当地轰动一时的“红延伸行动”（Operation Red Reach）中担任卧底调查员，该行动是一场涉及跨辖区、帮派关联毒品犯罪的重大复杂战役。面对侦破陷入僵局的困境，布巴尔展现了罕见的勇气，最终促成案件侦破并重塑了康特拉科斯塔县（Contra Costa County）的公共安全局势。团队没有盲目推销产品，而是直接致电布巴尔，诚恳地表示：“我们懂的其实不多，或许能帮上一点忙，但也说不准。我们能否登门请教，向您学习，理解您过往办案的卓越经验，然后再看看能否逐步做出有用的东西？”正是这种毫不自大、以极高信任为起点的谦逊姿态，让布巴尔和圣巴勃罗警局决定给这两个年轻人一次机会。 - **关键节点**：2018 年 2 月 26 日，团队正式入驻圣巴勃罗警局办公室并取得数据权限。 - **前期碰壁**：在获得第一张准入通行证前，遭遇了 20 多家机构的直接拒绝。 - **破局关键人物**：布赖恩·布巴尔（Brian Bubar），圣巴勃罗警局指挥官、前“红延伸行动”（Operation Red Reach）核心卧底调查员。 --- ### `13:00` 现场跟车体验与建立警务信任的鸿沟 在产品研发的最早调研阶段，团队就深刻体会到走出象牙塔、亲历现场的重要性。他们甚至在圣巴勃罗警局项目之前，就争取到了奥克兰警察局（Oakland PD）的夜间巡逻跟车（Ride-Along）机会。对于习惯了坐在办公室里的硅谷技术人员而言，通宵随警车穿梭在城市街头是一次极其震撼且大开眼界的经历。 彼时，通常会申请警车跟车的多为调查记者或社区权益倡导团体，硅谷工程师的身影极为罕见。警局虽然给予了礼貌周到的“白手套式”接待，但也让团队清晰地意识到：科技圈与基层执法机构之间存在着一道极难逾越的信任鸿沟。想要真正渗透并赢得这类机构的信任，绝不可能靠硅谷那一套快节奏的空泛宣讲，唯有沉下心来，投入漫长的时间与真诚。 --- ### `14:19` 在“削减警费”浪潮与极度极化中逆流前行 公司刚起步不久，美国便爆发了全美范围的“削减警费”（Defund the Police）抗议浪潮，随后又叠加上了 COVID-19 疫情与深刻的社会动荡。面对撕裂的舆论环境，创始团队从自身的认知底层出发审视公司的价值观：他们崇尚基于亲身实践（tactile/bottom-up）的认知方式，热衷于在行动中探索事实真相并质疑既定假设，且绝非出于愤世嫉俗的低信任心态，而是源于对复杂事实的敬畏。 在当时，创始人内心对那些“容易被外界误解的人群”有着天然的共情，而前线执法人员在当时恰恰处于这种被符号化、被严重误解的境地。这种社会现实造就了极其撕裂的日常戏剧感：团队在斯坦福等高校的昔日同窗好友正聚集在旧金山妇女中心门外高呼口号抗议警察；而团队成员每天清晨却开着一辆本田雅阁（Honda Accord），驶过海湾大桥离开旧金山，直奔警局与调查帮派凶杀案的重案组刑警并肩作战。团队同时包容并面对着这两种截然不同却真实存在的现实： > “站在泳池外面围观困局并不算什么本领，能够亲自跳进水里游过深水区，才是一种真正的特权。” 通过深入一线，他们发现许多被外界简单贴上对错标签的事物，内里其实充满错综复杂的灰色地带，问题越往下探究就越显立体与微妙。虽然经历了极度艰难的时期——当时技术招聘极度受阻，许多硅谷工程师一听到客户是警察局甚至连面试邀约都不予回复——但他们始终坚信，城市的终极目标是安全与繁荣。 Peregrine 致力于打造不偏不倚的“非政治化”（apolitical）基础设施，帮助服务公众的机构解决实际问题。如今，Peregrine 的业务已拓展至消防、应急救援、医疗卫生等多元公共安全领域，事实证明这种专注务实价值的道路并非不可调和的矛盾。 ```mermaid flowchart TD accTitle: Peregrine 穿越社会极化时期的双重现实 accDescr: 展示创始人团队在社会动荡期如何平衡两种认知，并提炼为客观务实的非政治化产品内核。 subgraph SocietalContext["社会撕裂与极化背景"] Protest["旧金山湾区：同窗好友参与反警察抗议"] Investigation["东湾警局：驱车与凶杀案侦探并肩解决案件"] end subgraph PhilosophicalMindset["认知底层与务实行动"] GroundTruth["拒绝盲目预设与冷嘲热讽<br/>扎根基层实操建立同理心"] ComplexReality["直面复杂与多维度的灰色地带"] end subgraph ProductMission["产品定位与最终愿景"] ApoliticalTech["打造非政治化（Apolitical）的基础设施"] CityProsperity["跨部门赋能：公安、消防、应急管理与急救卫生"] end Protest --> GroundTruth Investigation --> GroundTruth GroundTruth --> ComplexReality ComplexReality --> ApoliticalTech ApoliticalTech --> CityProsperity ``` --- ### `18:16` 颠覆传统模式：不做数据采集，专注数据打通与治理 在商业模式与技术架构上，Peregrine 与 Flock Safety 或 Axon 等传统公共安全软硬件巨头有着本质区别。传统的公共安全科技企业几乎全部建立在“数据采集”（Data Collection）的基础之上： 1. **硬件/传感器驱动**：向客户销售摄像头、传感器或记录设备，以被动捕获或人工录入的方式占有并存储海量数据； 2. **商业模式绑定**：依靠在客户系统内囤积独占数据和既有的销售分销优势，向机构兜售更多的采集工具和存储服务，其商业动力是无止境地扩大数据采集规模。 Peregrine 则彻底“倒置”了这一行之数十年的商业模式。Peregrine 坚持**不安装传感器、不引入新硬件、不制造新的冗余数据**，而是作为一个高度安全且受控的数据治理层，轻量地架设在客户现有的异构信息系统之上。 其核心业务是连接并打通各个孤立系统中本就属于城市与机构自身的数据，消除信息孤岛，从而帮助分析人员和一线决策者以极高的精确度和准确性快速检索、分析线索。Peregrine 的利益诉求与传统硬件厂商截然相反：它不靠无限制地搜集数据牟利，而是专注于提升人类与其自有数据进行交互时的精准度与安全边界。 | 对比维度 | 传统公共安全科技公司（如 Axon、Flock） | Peregrine | |:--- |:--- |:--- | | **核心商业引擎** | 依赖硬件/传感器持续扩张数据采集与存储 | 颠覆采集模式，专注现有异构系统的数据连接与融合 | | **部署形式** | 部署摄像头、传感器硬件及录入前端 | 纯软件/平台治理层，架设在现有基础设施之上 | | **商业动力（Incentive）** | 追求采集体量最大化以捆绑销售更多硬件 | 追求人机交互与检索分析的高精准度、治理严格性与安全性 | | **数据主权归属** | 往往将客户绑定在自身封闭生态与存储中 | 坚持城市自身的数据主权，不引入新数据，仅做分析与赋能 | ### `20:00` 数据主权哲学：破解公共安全与“监控国家”的悖论 在硅谷主流的数据商业模式中，大多数公司极度痴迷于通过汇聚用户数据来打造垄断性的“数据网络效应”。Nick Noone 与 Ben Rudolph 对此明确表达了反对立场。公众与社会各界对中央集权式的数据聚合体充满警惕——无论是公立部门还是私人科技巨头，一旦掌握了海量高特权敏感信息，极容易滑向滥用权力的深渊。Peregrine 选择了一条完全反向的去中心化路径：公司绝不以收集客户本没有的数据谋利，其核心使命是帮助机构安全、可信地用好自己手中原本就拥有、却受困于孤岛而无法调用的数据。 从创立第一天起，Peregrine 的底层架构便围绕“权限控制（permission controls）、数据治理（data governance）和数据主权（data sovereignty）”展开。Ben 强调，在关乎生死的高风险机构中，信任是不可逾越的红线。公职人员在为社区服务，数据本质上属于该机构与社区本身，绝不属于 Peregrine。许多人担心提升执法效率会导致建立“监控国家”，而 Peregrine 给出的答案正是将数据所有权与控制权彻底锁定在本地机构手中。在没有这类精细化治理工具之前，现实中甚至会出现工作人员“将一堆敏感数据刻录后塞进汽车后备箱、开着车穿越整个城市去交付”的荒唐场景，那种原始的数据流转不仅效率极低，而且毫无审计与安全保障。Peregrine 提供了细粒度的控制能力，确保机构在跨部门或跨层级协作时，既能按需安全共享特定信息，又绝不会导致过度共享（oversharing）。 - **数据去中心化**：不搞中央数据池，不靠劫持客户数据形成护城河，让每个机构完全保留对其数据的主权。 - **公共信任机制**：将产品定位为连接公职人员与社区信任的纽带，用高强度的访问控制消除公众对监控资本主义的恐慌。 - **安全共享替代物理外运**：以现代数据治理机制取代过去不可控的人工传递方式，消除后备箱运送数据等重大安全漏洞。 --- ### `22:57` 从基础搜索到深度因果分析：佛罗里达水上救援案例 谈及生成式 AI 在一线公共安全机构的落地，Ben Rudolph 坦言，初期的应用往往停留在比较浅层的“好用的搜索（nice search）”。过去警员可能需要在海量行列表格中苦苦肉眼筛查特定地址，而引入初始 AI 后，系统能把杂乱记录清晰地聚合成关联视图。但随着用户对工具理解的加深，整个部门的分析底线被迅速拉高，系统开始展现出以往单凭人力“根本无法完成”的深度归因分析能力。 Ben 分享了一个发生在佛罗里达州某县的真实案例：该县在短短一个月内执行了 100 多起水上救援（water rescues）——救援人员必须在洪水中驾驶冲锋舟，营救被困在熄火车辆或被淹民房中的市民。面对骤增的救援任务，指挥部门非常困惑，因为历史同期从未出现过如此密集的水难事件。他们开始在 Peregrine 平台中向 AI 提出疑问，平台背后的分析智能体（Agent）调取并融合了全县的历史灾情、911 报警记录、气象动态和县域专属的高精度地形数据，展开了多轮迭代推理。 ```mermaid flowchart TD accTitle: 佛罗里达水上救援突增的 AI 因果归因链路 accDescr: 展示气象、地理和接警数据在多轮推理后定位离岸流沙槽成因的分析链路 subgraph DataPrep["多源数据编排接入"] D1["911 报警与灾情数据"] D2["气象与风暴历史记录"] D3["县域高精度地形与沙滩地图"] end subgraph DeepInference["智能体多轮深度推理"] I1["对比历次类似风暴气象特征"] I2["发现该天气模式罕见地连续持续 3 天"] I3["持续风暴侵蚀海底形成异常沙槽"] end subgraph RootCause["可落地的因果结论与应对"] C1["沙槽引发致命离岸流 (Rip Currents)"] C2["向指挥官输出可落地的海域巡防策略"] end D1 --> I1 D2 --> I1 D3 --> I2 I1 --> I2 I2 --> I3 I3 --> C1 C1 --> C2 ``` AI 最终揭示了一个隐藏在气象表象背后的关键物理因果链：虽然类似的风暴强度过去也曾发生，但该特定气象系统“此前从未连续停留在该区域长达三天”。这种罕见的连续侵蚀在海岸带海底硬生生冲刷出了几道特殊的沙槽（sand channels），而这些水下通道恰恰构成了致命离岸流（rip currents）的绝佳温床，进而导致大量涉水市民瞬间被卷入险境。这一由 AI 挖出的隐秘因果直接为应急响应部门提供了高度可落地的研判支撑。创始人感叹道，如果工程师只是待在硅谷总部凭空空想“飓风救援人员到底需要什么功能”，那是绝对不可能预见并开发出这种能力的。 --- ### `25:00` 语义搜索破局：犹太会堂威胁事件的隐蔽模式关联 随后，团队分享了另一个依靠语义理解破解传统侦查瓶颈的实战案例。某地一名警探当时正在调查一起针对犹太会堂（synagogue）的威胁案件，他迫切需要查证周边两座犹太会堂近期是否也曾遭遇过潜在的反犹仇恨言论或隐蔽威胁。 在传统的警务信息化系统中，警探通常只能依靠预设关键词进行机械匹配。然而仇恨犯罪与威胁恐吓的表述方式千差万别，犯罪分子很少使用固定的标准词汇，传统的全文索引与布尔搜索在此类非结构化文本前几乎完全失效，侦查人员极易漏掉关键线索。借助 Peregrine 的底层语义嵌入（Embeddings）与大语言模型推理，系统能够在没有预先设定精准词表的情况下，跨越数以万计的警情记录、自由文本笔录和报案详情，自动识别出具有潜在反犹意图的威胁模式，将散落各处的零散关联事件完整呈现在侦查员面前。 - **超越传统关键词匹配**：依靠自然语言语义理解，打破“查不准、搜不全”的结构化搜索局限。 - **关联隐藏风险**：自动捕捉跨区域、跨周期的隐蔽恶意言论与恐吓动作，提前防范恶性仇恨事件升级。 --- ### `26:30` 冰山之下：95% 的数据预处理与零边际成本的软件构建 面对 AI 展现出的亮眼能力，主持人和创始人共同探讨了其背后的技术实质。他们指出，当用户在输入框中向智能体提出一个复杂问题时，表面看是模型在快速作答，但“95% 的真正工程量实际上发生在用户敲下提问之前”。Peregrine 耗费了极其巨大的工程研发点数，专注于极为繁重的数据编排与准备工作：清洗海量异构数据、打通不同系统之间的本体关联、对齐上下文，确保智能体在回答时不仅能做到准确无误，更能给出精准的溯源引用（citations）。 与此同时，AI 正在彻底重塑 Peregrine 内部的软件工程与交付模式。当代码生成的边际成本近乎归零时，交付模式发生了根本转变： - **平台化赋能一线**：Peregrine 核心平台向现场部署团队（Deployment Strategists）提供严格的安全边界、数据治理控制规范以及丰富的 API 接口。 - **量身定制微型应用**：并非每位客户都需要通用的分析 Agent，有些一线机构面临特定自然灾害时，可能需要的是一个轻量级的“飓风模拟器（hurricane simulator）”。部署专家能够借助 AI，在几天甚至几小时内直接在客户环境中快速生成这些高度契合需求的软件。 - **能力下沉至终端客户**：目前多数定制化开发主要由部署团队主导推进，但越来越多具备技术素养的客户也开始直接在平台上自行将业务构想转化为可运行的应用程序。 Ben 总结道，Peregrine 的技术演进是极具“涌现性（emergent）”的。公司的核心价值始终是对执法、消防、急救（EMT）等一线公职人员使命的深刻理解与同理心。无论前端技术展现形式是从早期的可视化图表、跃迁至当下的自主 Agent，还是未来出现的某种新型形态，“如果缺乏底层高质量、高度精确的数据源输入，哪怕做出一张花哨的图表或一个看似聪明的 Agent，在实战中也完全一文不值”。在 AI 席卷一切的新时代，Peregrine 紧抓数据质量与主权的核心战略从未动摇，而 AI 不过是用于放大一线人员使命成果的强大新杠杆。 ### `30:00` 长周期 Agent：以自动化重塑数据接入与集成 在探讨背景 Agent（Background Agent）和长周期 Agent（Long-horizon Agent）的应用时，Ben Rudolph 指出，Peregrine 平台约有 50% 的工程研发精力都集中在底层数据平台开发上，旨在让部署策略团队（Deployment Strategy Team）能够高效集成客户已经拥有的专有数据。截至目前，Peregrine 已经在各个客户体系中累计集成了数以万计的数据集。为了解决这项庞大、繁杂且不光鲜的工作，Peregrine 自研了面向集成的长周期 Agent，使数据接入走向自动化与智能化。 团队构建了极其严格的评测集（Eval Set），能够对 Agent 编写的集成脚本进行确定性的完整性与正确性评估。目前，Peregrine 约 90% 的数据集成代码都是由 Agent 自主编写完成，仅需部署团队在关键节点进行人工监督。这些 Agent 在后台往往持续运行数个小时：它们会深入分析客户数据库的底层架构，理解业务本体（Ontology），推断需要打通的数据片段，并派生出多个子 Agent 协同工作，实时向总编排 Agent（Orchestrator Agent）汇报进展。这种工作逻辑非常接近用于大型代码库迁移的 Coding Agent，其最大的优势在于结果具备明确的可验证性（Verifiable），从而大幅降低了 Agent 执行复杂任务的容错成本。 ```mermaid flowchart TD accTitle: Peregrine 数据集成 Agent 架构 accDescr: 展示总编排 Agent 如何解析数据库与本体，并派生子 Agent 编写集成脚本，最终由人工部署团队监督与评测集验证 subgraph Orchestration ["Agent 编排与解析"] DB["客户专有数据库"] --> OA["总编排 Agent (Orchestrator)"] OA --> ONT["本体分析与数据建模"] end subgraph SubAgents ["并行子任务"] ONT --> SA1["子 Agent A: 数据清洗与转换"] ONT --> SA2["子 Agent B: 字段映射与绑定"] ONT --> SA3["子 Agent C: 管道脚本生成"] end subgraph Verification ["验证与落地"] SA1 & SA2 & SA3 --> Eval["确定性评测集 (Eval Set)"] Eval --> Rev["部署团队人工审查与监督"] Rev --> Prod["90% 集成代码投入生产"] end ``` ### `31:45` 悬案破案 Agent 与一线执法的降负赋能 在业务端应用方面，Peregrine 将 Agent 能力重点投向了一线执法人员的业务痛点，其打造的第一个端到端业务 Agent 便是悬案调查 Agent（Cold Case Agent）。重特大或陈年积案的调查往往面临令人窒息的数据量——单个案件动辄包含 200 到 300 GB 的非结构化证据材料，涵盖海量执法录像、音频监控、现场照片以及堆积如山的 PDF 文档。在不少传统警局里，调查人员不仅要面对满屋子堆满纸质卷宗的大号档案盒，还要处理贴在材料上的成百上千张光盘，逐一查阅耗时极其漫长。 Peregrine 延续了与一线用户共创产品的传统。最初，他们与一个曾成功洗刷冤屈、为一名被错误定罪男子平反的警局团队合作。对方提出一个挑战：能否让 AI Agent 复现刑警人工排查出的平反证据？Peregrine 将该案全部材料接入系统，经过长达 30 至 60 分钟的长程推理与深度分析，Agent 最终成功推导并复现了人类侦探耗费数月才得出的无罪结论。这一原型随后在全美多个地区推广部署。在威斯康星州某县的一起严重案件中，面对 300 GB 的混合证据，Agent 在数以万计的杂乱数据中精准抓取到了几条关键的蜂窝基站通话详细记录（CDR），不仅证实嫌疑人出现在犯罪现场，还将其实时定位到了最终弃尸地点。Nick Noone 强调，Peregrine 在取得此类重大成果时始终坚持扮演“幕后的安静专家”（Quiet Professionals），绝不向外界高调宣扬功绩来抢一线警察的功劳，因为在公共安全领域，大肆吹嘘只会迅速摧毁好不容易建立的信任。 - **海量非结构化证据**：重案分析通常需要处理 200–300 GB 的音视频、图像与扫描文档，耗费刑警数千小时的人工视听时间。 - **确定性复现**：通过 30–60 分钟的长周期逻辑推理，Agent 成功重现了人类警探长达数月的冤错案件平反论证路径。 - **关键细节挖掘**：在威斯康星州案例中，系统在浩繁证据池中挖掘出散落的手机基站 Ping 点记录，一举锁定案发地与抛尸现场。 - **低调务实的伙伴定位**：坚持让一线公职人员站在聚光灯下，拒绝将客户案例当作公关噱头，以维护极其脆弱的警民与政企信任。 ### `35:02` “反网络效应”：维护数据主权与对抗全景式监控 随着公众对 AI 监视和“中央集权式监控”（Centralized Panopticon）的恐惧与日俱增，许多人担心所有数据会被集中到一个吞噬一切的利维坦系统之中。Nick Noone 坦言，这是一种深植于美国社会内部的不信任感，而许多追求增长的技术公司习惯走捷径：在灰色地带爬取开源数据或打擦边球，强行聚合跨部门数据来制造所谓强大的“网络效应”。 Peregrine 明确将自己定位为一种“反网络效应”（Anti-Network-Effect）业务。Nick 指出，他们坚决不打破法律、法令规章（Ordinances）或部门标准作业程序（SOP）。技术的核心使命绝不是把各机构的数据收缴整合进统一的监控网络，而是全力捍卫每一个独立部门的数据神圣性（Sanctity of the Data）与本地数据所有权。真正的技术挑战在于：如何在完全不侵犯各自主权的前提下，构建安全合规的数据治理模型、极度精细的权限控制逻辑（Permissioning Logic）与可互操作的连接纽带。对 Peregrine 而言，企业经营是一门“学会放手”（Practice in letting go）的修行——克制追求盲目扩张的冲动，用最高级别的透明度与严谨治理，消除民众对技术滥用的顾虑。 ### `38:40` 人脸识别的道德抉择：拒绝硅谷式的傲慢规训 当被问及在面对如“人脸识别”（Facial Recognition）这类极具道德争议的技术时如何抉择、遵循怎样的“北极星”准则时，Nick Noone 直言不讳地批评了硅谷科技公司的傲慢态度。他认为，一家硅谷公司试图越俎代庖，替整个公共安全行业甚至整个社会做价值预设或推行一刀切的通用决策，在逻辑和道义上是根本错误的。 Peregrine 绝不会以技术权威的姿态向机构发号施令，强制要求对方采用某种特定算法或遵从特定的数据保留期限。相反，技术公司的职责是保持极大的耐心，帮助客户全面看清所处环境的法律、技术与社会现实，揭示他们可能尚未意识到的潜在风险与约束，引导其选择最契合本地社群意愿的数字化路径。事实上，全美绝大多数公共安全机构及其代表的本地社区，都在基于自身制度考量后主动选择不部署人脸识别。尊重各社区和机构的主权决定，远比由技术精英自上而下灌输意识形态更为重要。 ### `40:00` 为弱势群体打造技术：打破八位数合同的规模化神话 访谈深入探讨了如何为长期被高科技遗忘的“弱势群体”（underdogs）提供技术支持。主持人指出，像 Palantir 这样以部署工程（Forward-Deployed Engineering）闻名的企业，传统上只接八位数（千万美元级）起步的巨额订单，而外界普遍认为在资源匮乏的地方机构和中小城市推行这种深度的客户服务模式，其经济模型根本无法规模化。对此，Nick Noone 和 Ben Rudolph 直言，这需要极大的自信与魄力——去深耕并构建一整套垂直整合的开放技术栈（包含网络访问、权限治理、ETL、数据流水线、本体论（Ontology）、前端 UX 与 API），不仅在每个模块上做到一流标准，更要把曾经昂贵无比的解决方案降维提供给中小城市客户。 Peregrine 坚信“跟随客户、遵循法律”的原则，不在脱离具体场景的情况下武断设定技术的红线。各个城市之间经常互相打电话咨询建议与最佳实践，Peregrine 的定位是为它们消除信息壁垒、理清左右边界与机构规范。在实施层面，Peregrine 愿意耗费时间陪客户“一路走到最终结果”（all the way to the outcome），哪怕初期看似极度人工、笨拙且不稳定。正是这种亲自下场的重度投入，反倒倒逼公司打磨出了既能深入基层解决具体痛点、又能随着时间推移大规模扩展的底层系统。 - **全栈垂直整合构架**：Peregrine 从底层网络与治理权限、ETL 流水线、本体系统一直自研到上层应用界面与 API，确保系统在保持开放可互操作的同时，满足严格的合规标准。 - **敢为天下先的经济模型**：打破只有军工或联邦巨头才能负担起定制化数据智能平台的惯例，使中小地方政府与公共安全机构以低门槛享受到顶级技术。 - **与客户共担责任**：不只卖软件授权，而是亲自下场陪同客户攻克业务闭环，将“看似不可规模化”的手工作业提炼为通用的平台能力。 --- ### `42:54` 信任一线贡献者：“黑暗洞穴”里的部署工程师文化 对于外界盛赞 Peregrine 部署工程师（Deployment Strategists / Forward Deployed Engineers, FDE）不仅技术过硬、极度谦逊，更深度融入了服务公共事业的使命感，Nick 分享了公司的管理理念。Peregrine 内部推崇高强度的追求卓越与直面问题的反馈文化，其核心支柱在于对一线独立贡献者（IC）的极致信任。在传统企业将现场部署视作“成本中心”（Cost Center）的背景下，Peregrine 旗帜鲜明地将其定义为驱动创新和收入扩张的“研发中心与增长引擎”。 ```mermaid flowchart TD accTitle: 一线边缘创新反馈闭环 accDescr: 部署工程师在一线发现需求、快速原型，并向核心研发输送通用技术原语的飞轮架构 subgraph Fringe ["现场前线：部署工程师 (FDE)"] A["深入一线客户业务场景"] --> B["克服技术限制完成任务 (Hacky Prototyping)"] end subgraph Signal ["双向创新反馈"] B --> C["沉淀技术原语 (Primitives)"] B --> D["创新实验室方案 (Custom Solutions)"] end subgraph Core ["平台底层研发 (Core R&D)"] C --> E["转化为平台核心通用功能"] E --> F["赋能下一轮现场交付与扩张"] F --> A end ``` Nick 用“黑暗洞穴”作比喻来阐释一线自治：公司派工程师腰别工具带深入漆黑的未知洞穴，洞口有一根救援绳索和可以呼救的伙伴，但走入深处后，全凭工程师自身的正直品格与原则去独立决断、创造价值。Peregrine 深信真正的创新发生在组织最边缘的前沿地带，产品、工程与设计团队的灵感正是由遍布全国乃至跨国驻场的现场团队带回来的。 - **黑暗洞穴隐喻**：给一线工程师最趁手的工具与安全绳，随后给予充分授权，靠职业操守与原则在技术荒野中探索攻坚。 - **内部增量试点模式**：落地客户后，现场团队以快速冲刺方式持续开辟新用例，实现“以点带面”的二次扩张。 - **现场倒逼研发的记忆**：Ben 回忆创业初期 Nick 开车过旧金山海湾大桥前往圣巴勃罗警察局（San Pablo PD）时，自己在副驾上疯狂敲代码，只为赶在与侦探 Aaron Blaisdell 碰面前上线对方迫切需要的报表功能，毫无空洞务虚的说辞。 --- ### `45:00` 现场创新的两类产物：从“土法编辑”到飓风模拟器 Ben 将前线部署工程师催生的产品创新分为两大阵营。第一类是**核心技术原语（Product Primitives）**。Peregrine 曾长期缺少对特定字段和属性直接编辑的功能，一名部署工程师为了帮客户解决问题，“绕路”写了一个解析脚本：让用户在对象下发表评论，系统自动读取评论内容并反向刷新对应属性。管理层见状立刻意识到底层迫切需要官方编辑能力并迅速跟进。Ben 强调，前线的核心考核是“达成目标，管它技术受不受限”（your job is to hit the objective, like to hell with the technology），这种看似粗糙甚至 hacky 的自发解法，正是检验真实需求的最佳信号源和快速原型。 第二类则是如同**创新实验室（Innovation Lab）**般的专项定制方案。这些功能过于特定，可能永远不会被并入通用产品，却展现了前线极其惊人的创造力： - **飓风模拟器（Hurricane Simulator）**：某位部署工程师利用 Peregrine 的现有数据流搭建了防灾抗灾的飓风态势演练沙盘，连创始人看后都惊呼“我甚至不知道平台还能这么用”。 - **消防局选址规划模型**：另一位入职仅一个月的女性部署策略师，整合了全市历史 911 报警呼叫响应时间、市政预算等多源异构数据，允许决策者在地图上任意放置规划中的新消防站，并实时推算其能辐射改善的人口与救援覆盖面。 --- ### `48:50` 迈向一万座城市的基石：让每一个独特的城市变得更酷 展望十年后的终局愿景，主持人提问：若 Peregrine 顺利成为全球 10,000 座城市的“机构记忆层”（Institutional Memory Layer），面对跨越政治周期的巨大权力与相伴而来的责任，团队如何自处？Nick 回答，10,000 座城市的终局要求公司在本质上将自身定义为纯粹的“技术基础设施提供商”——赋予城市用好其自身数据的权利与能力，而非公司自命不凡地对城市施加权威。“每一座城市、每一个司法管辖区都是独一无二的，保护这种独特性本身就无比美好”，Peregrine 的使命就是让这片各具特色的百宝箱（potpourri of cities）以属于它们自己的方式变得更棒（make each one more awesome）。 通往目标的路径不能仅靠口号，而必须体现在每一个行动细节与道德指南针上。在以往，大型科技公司在服务州、县、市级公共安全部门时留下了长串的失败记录，根本原因在于中小机构无法承受高昂的定制交付成本。Peregrine 借助前沿技术与高效率工程，将服务成本降低了整整一到两个数量级（跌破每年百万美元门槛），这才赢得了服务广大基层的入场券。面对技术权力，团队抱持着极高的敬畏心与管家精神（stewardship）：数据与决策的真正权力永远属于公共机构本身，而Peregrine 唯有以严谨的审慎态度，继续证明人工智能可以切实守护社区、造福民生。 - **基础设施的定位**：坚持只做赋能的中立底座，尊重不同城市的法律与制度特色，保留各地方社区的治理主权。 - **量级维度的成本革命**：将原本动辄千万级的高门槛解决方案降维至中小城市可支配预算之内，打破该领域多年来巨头接连折戟的怪圈。 - **科技向善的管家责任**：践行严肃谨慎的数据伦理，树立以 AI 实质性改善社会安全与公众生活的工业级标杆。 ## Notable quotes > `02:07` 为客户、并代表客户一路挺进到最终成果（getting all the way to the outcome） — Nick Noone > `03:58` 放下自己的聪明才智、放下自身的专业光环、悬置个人的优越感（suspend our ego） — Nick Noone > `05:23` 大量看似无解的现实社会困境，追根溯源都是“下游数据问题（downstream data problems）” — Ben Rudolph > `10:00` 我们懂的其实不多，或许能帮上一点忙，但也说不准。我们能否登门请教，向您学习，理解您过往办案的卓越经验，然后再看看能否逐步做出有用的东西？ — Peregrine 创始团队 > `14:19` 站在泳池外面围观困局并不算什么本领，能够亲自跳进水里游过深水区，才是一种真正的特权。 — Peregrine 创始团队 > `26:30` 95% 的真正工程量实际上发生在用户敲下提问之前 — Ben Rudolph > `26:30` 如果缺乏底层高质量、高度精确的数据源输入，哪怕做出一张花哨的图表或一个看似聪明的 Agent，在实战中也完全一文不值 — Ben Rudolph > `31:45` 幕后的安静专家（Quiet Professionals） — Nick Noone > `35:02` 学会放手（Practice in letting go） — Nick Noone > `45:00` 达成目标，管它技术受不受限（your job is to hit the objective, like to hell with the technology） — Ben Rudolph > `48:50` 每一座城市、每一个司法管辖区都是独一无二的，保护这种独特性本身就无比美好 — Nick Noone ## Mentioned Peregrine, Nick Noone, Ben Rudolph, Palantir, 美国特种作战司令部 (SOCOM), 洛杉矶警察局 (LAPD), 联合国难民署 (UNHCR), Dimagi, Airbnb, 圣巴勃罗警察局 (San Pablo PD), Brian Bubar, 红延伸行动 (Operation Red Reach), 奥克兰警察局 (Oakland PD), Flock Safety, Axon, Aaron Blaisdell, 康特拉科斯塔县 (Contra Costa County) ## Video description Most public safety technology companies grow by collecting more data. Peregrine inverted the model: no sensors, no new data, a business built on connecting the data and information cities already own. Co-founders Nick Noone and Ben Rudolph received more than two dozen no's before San Pablo PD let them in the door in February 2018. Today, Peregrine powers law enforcement, emergency medical services, fire and rescue, and other services in more than 400 cities and communities globally. Nick and Ben explain their north star for data sovereignty, and discuss how Peregrine's philosophy and privacy-first approach to data access and ownership preserves individual privacy and cities' sovereignty. They walk through how AI and long-horizon agents are being deployed: a cold case agent that reproduced an exoneration detectives had reached by hand, a Wisconsin county that placed a suspect using cell records buried in 300GB of evidence, identifying threats to a synagogue, root-causing an escalation in weather-related incidents, and more. 00:00 Introduction 02:07 What Forward Deployed Engineering Means 03:58 What Silicon Valley Gets Wrong 05:23 UNHCR, Dimagi And Downstream Data Problems 08:25 Why Cities, Why Safety 10:45 Two Dozen Nos And San Pablo PD 14:19 Building Through Defund The Police 18:16 The Inversion Of The Collection Model 21:20 Data Ownership And Governance 22:57 From Nice Search To Deep Analysis 29:59 Agents Writing The Integrations 31:45 The Cold Case Agent 35:02 The Anti-Network-Effect Proposition 38:40 Facial Recognition And Hard Decisions 40:48 Technology For The Underdogs 42:54 Trusting The Individual Contributor 48:50 Ten Thousand Cities [Source](https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI4167193786.mp3)

[2:07] Nick Noone: 为客户、并代表客户一路挺进到最终成果（getting all the way to the outcome） [3:58] Nick Noone: 放下自己的聪明才智、放下自身的专业光环、悬置个人的优越感（suspend our ego） [5:23] Ben Rudolph: 大量看似无解的现实社会困境，追根溯源都是“下游数据问题（downstream data problems）” [10:00] Peregrine 创始团队: 我们懂的其实不多，或许能帮上一点忙，但也说不准。我们能否登门请教，向您学习，理解您过往办案的卓越经验，然后再看看能否逐步做出有用的东西？ [14:19] Peregrine 创始团队: 站在泳池外面围观困局并不算什么本领，能够亲自跳进水里游过深水区，才是一种真正的特权。 [26:30] Ben Rudolph: 95% 的真正工程量实际上发生在用户敲下提问之前 [26:30] Ben Rudolph: 如果缺乏底层高质量、高度精确的数据源输入，哪怕做出一张花哨的图表或一个看似聪明的 Agent，在实战中也完全一文不值 [31:45] Nick Noone: 幕后的安静专家（Quiet Professionals） [35:02] Nick Noone: 学会放手（Practice in letting go） [45:00] Ben Rudolph: 达成目标，管它技术受不受限（your job is to hit the objective, like to hell with the technology） [48:50] Nick Noone: 每一座城市、每一个司法管辖区都是独一无二的，保护这种独特性本身就无比美好

[0:00] Nick Noone: psychologically going in to own or co-own the problem [10:45] Nick Noone: Brian, we don't know that much... can we come in, ask you some questions, learn from you... and then maybe build something over time? [18:16] precision and accuracy in how a human being interacts with their data [35:02] Ben Rudolph: We have to protect against those moments where it's not like we don't need to break the rules, nor obviously should any organization break the rules in order to find a way to introduce AI or technology into these complex environments. The opposite is true. We have to protect the data.

[0:00] Peregrine is really around this idea of how can we leverage technology to work with our cities, our counties, our states to impact the places that we live in? At the bottom of the pyramid, really, when it comes to how to make cities awesome, there's is the idea of safety. The idea that people... Need objective safety. and also need to feel safe. End. when that stability is there, it's amazing what's possible. And there are all these kind of contrasting but but but actually similar concepts that we were wrestling with at the time. We were thinking about how to deploy inside of institutions and also find ways to eliminate the mishandling of information inside of really complex organizations that have access to super, super sensitive data. And then if we could create N. of the backbone infrastructure level of the modern city, then we could help to preserve individuality. We talk about privacy and other aspects of what it means to be like an individual person, but also bring people together. One of the most important stories in AI is from a company that you may not have heard of, but is almost certainly keeping you and your loved ones safe. Paragran is building AI that protects cities and communities, and they are rejecting the surveillance state while doing that. And this should be a fantastic conversation about one of the most important applications of AI, forward deployed engineering, civil liberties, and much more. So let's get into it. Nick, you ran Palantir's SOCOM unit. You deployed into dangerous high stakes intelligence operations in the Middle East. Tell me what does forward deployed engineering actually mean and what does Silicon Valley get wrong about it? Palantir was a really fun experience and a really formative one because before joining the company, I had no idea what this concept of forward deployed engineering was. meant But I was surrounded by people that were running into customer environments and... doing who knows what. Yeah. I watch them. I watch them leave the office. I watch them come back to the office and tell stories. Um, It's felt to me like the adventure of a lifetime to be able to go... into a customer context and own the customer's problem. but also... take pride in the fact that the problem belongs to the customer, and then take pride in getting the customer to the win. And so they're, These two... kind of these two truths that... I learned. when I was an early forward deployed engineer and that we try to embrace as we built our company. And those are the We've... psychologically go in and own or co-own the problem. We talk internally about getting all the way to the outcome. with and on behalf of our customers. Technology and all of the skills and ways of delivering our Tack. is part of the answer, but the real answer is just getting to the outcome at all costs, ideally three to five times faster than any other person or team could do. and everything flows from there. But recognizing that at the end of the day it's the customer's win. It's not our win. And taking deep pride in the impact that we have on the institution and on the people that we're sitting across the table from. That's where the magic is culturally. What does Silicon Valley get wrong about it? When we first started building Peregrine, The forward-to-play engineering concept had been around for a while, and we got feedback that when really smart people from top five, top 10 or equivalent schools would go into a really complicated environment. place like the LAPD. End. quickly try to understand 30 years of institutional context. not just at the data level, but at the human level. How does this organization work? How are decisions made? How do teams interoperate with each other? The idea that, moving fast is even possible in that environment is very counterintuitive to the customer. And unfortunately, I think that shows up as ego to a lot of people. organizations where People have dedicated their entire lives to a particular role. And I think Silicon Valley... [5:00] misses that sometimes that letting go of our intelligence. letting go of our own skills and abilities. Trying to. suspend our ego. and then really get into the customer's context. It's an easy thing to say, but it's such a high empathy patient. way of working. Yeah. I love that. Ben, while Nick was off deploying into the Middle East, you were in almost the opposite part of the world in every sense of the word. You were doing refugee work. in Africa, in India. But you landed on a similar thesis of... the world and technology's role in it. Maybe say a word on that. Yeah, absolutely. So I, For me, I've always been obsessed with this idea of... How do you make impact with technology? I graduated school and I remember I had these two offers. One was to go work at a small startup at the time, Airbnb, and the UN Refugee Agency, which is an organization that goes and helps refugees who are making these extremely dangerous treks across country lines, usually because of war, to try and get help. took that path and deployed to the Sudanese border, the Colombian border, and was deeply humbled and really understood deeply understood these types of situations and where where technology can help and where honestly it can't help. And I remember... The thing that I took away from that experience is UNHCR is this, is this, deeply disconnected organization. It does phenomenal work, but it's deeply disconnected. They have some data. It's often in spreadsheets, and it's very difficult to actually make sense of. And so that's where I started to form this thesis of a lot of these problems are downstream of data problems where you can have a lot of impact. After UNHCR, the UN Refugee Agency, I left to join an organization that you probably haven't heard of. It's called Demagi. They're a small company. They work building last mile healthcare solutions in under-resourced places around the world. And so like one specific example, one project that I worked on while I was there, we were working with the Indian government on this project, building applications to help people, folks in rural India adhere to their tuberculosis drugs. And at the end of the day, we did get it deployed. And it's, That technology rapidly spread throughout the country. And that was a really cool moment for me and really formed my thesis and idea around how Peregrine can have impact and how you can, how technology can have impact in, uh, in the communities that, that we live in. And so Peregrine is really around this idea of how can we leverage technology to work with our cities, our counties, our states to impact the places that we live in. Take us back to 2017. You cold called your way into the San Pablo police department. What was going through your heads when you did that? In 2016, when Ben and I... got together actually on a vacation with a bunch of our gymnastics teammates. We had a conversation, basically 10 years after we graduated saying, should we build something together we got together in san francisco and with My experience in the national security space and similar in the U.S. government is, Ben's experience working on humanitarian causes and On my end, I'd been flying back and forth between Baghdad and similar places in the Middle East. Ben was commuting God knows how many flights into Africa. We looked at ourselves and said, this is the place where people live. This is the place where people thrive. This is where the majority of our lives are spent. And then we started thinking about how to deconstruct working with American cities. And what we came to is... At the bottom of the pyramid, really, when it comes to how to make cities awesome, I'm is the idea of safety. The idea that people need objective safety. and also need to feel safe. End. when that stability is there, it's amazing what's possible. And there are all these kind of contrasting but but but actually similar concepts that we were wrestling with at the time. We were thinking about how to deploy inside of institutions and also find ways to [10:00] Eliminate. the mishandling of information inside of really complex... organizations that have access to super, super sensitive data. And then if we could create n. of the backbone. infrastructure level of the modern city, then we could help to preserve individuality We talk about privacy and other aspects of what it means to be like an individual person, but also bring people together. End. The idea that we could do both of those things, preserve and protect individual privacy while also bringing people's people together and help them realize that if we're staring at the same information, we kind of want the same thing. Yeah. Like, That's what I find really beautiful about cities. And I think we're very aligned on that. How many people said no before San Pablo said yes? It's such a blur at this point, but it's definitely over two dozen. In my opinion, we started the company on February 26th of 2018. We'd incorporated the company a month or so earlier. But February 26th is the day that the San Pablo, Northern California Police Department, permitted us to step into the building with our badges we had desk space and we had access to information so that we could actually get to work. And the reason we got there is, we started to, research subject matter experts in the art of public safety at the city scale, at the municipal level. And we found this amazing article about a young... up-and-coming commander in a local police department named Brian Bubar, who was catapulting up the ranks inside of a paramilitary organization that's very structured, that takes years and years to reach higher levels of leadership. And he was young and intelligent and creative. And there's this article we found online about Operation Red Reach, which was one of the most iconic characters. cross-jurisdictional gang-affiliated narcotics. investigative you know work uh that happened in northern california at the time um And Brian was a covert... Operator. trying to figure out how to move this stuck investigation forward and I thought that took so much courage for someone like Brian to come in And I, be the reason Operation Red Reach, and it obviously wasn't just Brian, but so many people surrounded that work. and did something so impactful. and made headlines but more importantly you know, changed the state of safety in Contra Costa County. We just called him. and said, Brian. We don't know that much. Hmm. We may have... some utility. but I don't know. Can we come in ask you some questions, learn from you, Try to understand some of the awesome things you've done, including some of these amazing prior investigations. Yeah. And then maybe build something over time and just see what happens and lead with that level of trust. And that was what finally got us access. It's like a chance on you. Yeah, truly. Yeah. Mm-hmm. Do you do ride-alongs? We have amazing photos from those days. Our first ride along was actually with Oakland PD. Yeah. Which was which was a way when we were still in this like deep research. And that was a really eye opening and interesting experience. Overnight, I think. Yeah. And it was also eye opening. It's so I mean, the most people that are coming, I think, from from Silicon Valley going on ride alongs are not necessarily engineers at the time, you know, reporters or other other types of, you know, community interest organizations. And so, you know, I think we we had the white glove service at the time. Yeah. And it really was it was eye opening that it's so difficult to penetrate and build trust with these organizations. that signaled to us that we just have to take our time. Yeah. You started this company... 2017, 2018. I think. defund the police was, you know, was starting to get going maybe in the early days of the company. What did that feel like? And how did it make you dig deep and think about your values? I'll share that, you know, on a personal level, I think I'm a very, I tend to be a very experience-based learner, like very tactile, very, you know, want to hear the concepts, but very, build a bottoms up understanding through action and, and, I have found in my life with that mentality, it fundamentally leads me to question assumptions. And... [15:00] Um, The intent is not to question assumptions in a low trust way, in a skeptical way at all times. It's not, you know, I think there's a danger of becoming overly cynical, but... to really understand truth and love like curious adventure um I think that is... the kind of that, the psychological... backing that led us to build our business in the midst of a really tumultuous time in America. You know, I have a personal affinity for working with People who... sometimes feel misunderstood, frankly. And I'm, Um, it just, it just is, it's, it's, it's a, it's a very intuitive passion for me, um, to help, help, help them feel more understood. Uh, I think many people at our company probably have, have character qualities like that, frankly, um, but. And, and, When I think about moments like this, COVID and... the protests that were happening at the same time, the convergence of social unease in America. We were holding a lot of complexity. We had very close friends that we've graduated with and built things with in the past that were standing outside of the Women's Center in San Francisco protesting the police. and We were driving a Honda Accord across the Bay Bridge out of San Francisco every day to work with gang homicide investigators. And we were holding those two truths at the same time. And... I... To me, It's not enough to... to look at a very difficult situation from the outside like the privilege is jumping into the pool and swimming um And I think through that process, we establish empathy. We realize that, oh, this, you know, this, this thing that we thought was wrong is actually really complicated. There are a lot of gray zones. And actually, the deeper we go, the more nuanced and textured the problem actually is. How empowering is that? right um And, And so in some ways, even though in that period of time, our business is much more diverse now, right? I mean, we work like the through line is delivering safety and prosperity to cities. But At the time, we were only working with police departments. Now, police, fire, emergency management, health services, all kinds of stuff going on, but Engineers... wouldn't. even respond. When we made a recruiting call, I mean, it was a fascinating time. And I think that that level of polarization to me, for both of us in our lives, I think we've seen the pendulums shift and. we're trying to deliver solutions that are apolitical, that are what, people, communities, groups of people, these institutions that serve them truly want and need. And in many ways, I actually don't think that's controversial at all, if done well. How is your business structurally different from the data collection companies like a Flock or an Axon? Mm. Okay. So we've been operating in public safety since 2018. And I think most of the companies that have come before us have built their business fundamentally on the back of data collection. So the idea that a hardware or software solution is a good idea. is installed inside of a customer base and is fundamentally about the input, whether it's passive collection through a sensor or, you know, human beings inputting information directly, the collection and storage of that information. Historically, those companies have grown because once they have information that it belongs to the customer inside of their system, plus distribution advantage, they can kind of sell more stuff. And those, the additional stuff tends to be more collection systems. Peregrine is like the inversion of that entire model. We're fundamentally in the business of joining disparate information to provide a more secure, properly governed solution that sits on top of the pre-existing systems that helps people do better work, helps people get more precision and accuracy in the answers to their questions. I think for companies like ours, being not in the business of bringing more data to a customer, but in the business of... building solutions that drive greater levels of precision and how a human being interacts with their data is a kind of like flipping the historical model on its head, so to speak. Okay, so structurally almost the business incentive is the opposite almost. It's not about maximizing the amount of data that [20:00] you're collecting and creating network effects of that data. I think the world is very concerned about the idea of like the amalgamation of data and the kind of like central, you know, whether public sector or private sector, kind of authoritarian body that has access to this privileged information and what will they do with it? And for us, we almost want to decentralize that to get to a world where like we are actually not in any shape or form in the business of to the customer that they don't already have. The fundamental problem is that they can't utilize the data that they have or utilize it in a way that's secure and high trust for the communities that they serve. I'll just say from day one, we've been... really focus and thinking about this idea of permission controls, data governance, sovereignty. These are necessary ingredients to kind of deploy to these types of high stakes institutions. And Nick is right when he talks about trust. And we think a lot about trust, trust both with the community. These are the, And these public servants are ultimately serving the community members. We have trust with those public servants and the community that we have to uphold. And so we think a lot about this and how we build what we build and how it interacts with the end user. Yeah. It seems like the data governance and data being so locked down is almost a way to reconcile the tension between... public safety and not becoming the surveillance state. That is the answer. idea of you have local intelligence, local data, everything's locked down. Actually, maybe share a bit about your philosophy on data, data ownership, and data sharing. We deeply believe in the idea that... each. customer or each institution. owns their own data. This is ultimately the Thank you. That. organization is serving the community. That is the community's data. That is the organization's data. It's not Peregrine's data. And we think about providing the controls and capabilities to allow them to very securely manage. roll this out within their own department, but also share this with select pieces of information when they need to. Without these types of solutions, you get folks, you will put a bunch of data in the back of a car and drive it across a city. And there's actually less control there. And so we think a lot about where we're, Where can we actually enable people? the outcome for our user and provide the controls that are specific and robust that enable them to not overshare, share too much, and that they can trust in what they're actually doing. delivering. What are the biggest use cases you see for your customer base? Maybe what are people starting to play with now? And then where do you see that going? Yeah, this is what excites me the most. I'm I. Me too. Yeah, I think I think I think when you initially do deploy this type of AI to these organizations, you often just get what I'd say a nice search. You know, maybe you were looking for a dress and now you don't have to look at a bunch of rows and you see everything that happened out of the way. I think... As we have had users get more used to the product and understand it better and how it works, you start to see the floor get raised for everybody across the department. You start seeing these really interesting, deep types of analysis that previously were just impossible. So, I'll give a couple examples here because this stuff's really interesting to me. We're working with this County in Florida. And the other month they had to do over 100 water rescues. And they're like, why? We've never had to do this many water rescues before. Water rescues is when there's a flood... The rescuers need to get on a boat and literally go out and rescue someone from their car that might be stalled in the middle of the flood or their home. Yes. And so they started they started asking Peregrine and starting interrogating this question and. And what they started to uncover is that, okay, these weather patterns, they've happened before. But after a couple iterations and the agent doing some deep research, they're going to be a little bit different. it came out with the pattern that, oh, these weather patterns have never occurred for three consecutive days. And those types of weather patterns, they create these sand channels. And those sand channels are the perfect conditions for rip currents. And those rip currents cause sand. a lot of issues for people who are in the water at that time. And that is actionable for the agency. And they can think deeply about that. [25:00] And overlay their terrain maps and all of the unique things that are county-specific. bunch of like they obviously have to have incidents information 911 calls they have to have weather information all that information is integrated and then kind of at uh at the disposal of the ai to help the the the customer yeah and we just can't i mean there's no way we could have made this up right it's like we we were sitting in in our headquarters trying to pontificate on what you know emergency responders in a hurricane needed like there's zero chance we I mean, another one that I'm excited about, there was a detective who was investigating a threat against a synagogue. And the... uh what he was trying to do was trying to figure out were there any other anti-semitic threats that had happened to these two synagogues in the area now when you think about that question uh and you're a detective what are the keywords that you search for to try and find that information it's actually very difficult for for keywords to find that and what gets me uh excited this is another area that's just previously impossible and With the help of AI, they're able to semantically understand a lot of the information, all that they have access to, and they're able to actually pull out and they were able to find a bunch of pattern of these threats that were occurring against these synagogues. And I thought, you know, that was pretty interesting to me and a novel use case. Yeah. Interesting. Okay. So you got semantic search and embeddings going on. You got, sounds like reasoning models that uncovered the use case. Yeah. So one of the most interesting things is because you've brought in all this context and you've stitched it together together. You can kind of let the AI... Do its thing. Certainly 95% of the work is... what happens before the user types in the question. What is all the preparation that you do to get to a place where the AI can act? answer accurately, cite those questions accurately. That's a really hard problem. We spent a lot of points on that. We have a lot of agentic use cases and I, you know, A lot of our engineering points are spent on that data preparation, getting that AI ready. There's also this idea that AI allows you to write software rapidly. And what happens when the cost of generating the software is virtually zero? And what if we could apply... provide the platform for our deployment team that gives them the security and government's controls, that gives them the APIs they need, and then they're allowed to write the world with software. And so what we're seeing is that these deployment strategists, not every customer needs a, a, an agent, maybe they need a hurricane simulator. And the way they're able to write this really well, and is with is with AI. And I think this rapid innovation cycle is just super interesting to me. So, so cool. Is it your deployment strategists that are... kind of coding things in customer environments? Or do you think your customers might even, you know, they have all these ideas in their head of what they want? Can they go from idea, to working application themselves in your environment? 100% we enable that. I think the... you know, what we see most often is a partnership with our deployment team to, you know, to provide that kind of technical expertise and be able to take the idea and bring it to fruition. We do have some customers that are able to kind of leverage the platform in that way. But a lot of that motion comes with our deployment team. Do you think that'll change over the coming years? I see it changing right now. I mean, I'd love to get your thoughts on this, Ben. I think it takes confidence to embrace that and say, you know what, like, I'm going to Let's innovate and may the best method win. So much of what Peregrine does on the technology side is emergent. Peregrine has been providing domain expertise and analytics. for our customers. We deeply understand and empathize with law enforcement, fire departments, EMTs, And then we provide technology to enable them to achieve their most important missions. And that technology can change over time. And that's like the great thing about it. Most of that stuff is downstream of... really high quality data and you can build charts And that Chart could be completely useless if it's not. accessing the the right information and accurate information. The same is true with agents, and the same will be true with the next technology. And so I believe the core strategy of Peregrine has remained very unchanged in this new age of AI. And it's just we are applying a new technology to see how it impacts our customers. [30:00] Are you doing anything on the long horizon agent side or background agents side? Yeah, absolutely. A couple of things here. So I think stepping back a little bit, like what is the Peregrine platform? We think about 50% of our engineers, spend most of their time working on the data platform that enables our deployment team to integrate data that the customer already owns. Mm-hmm. And we've, at this point, integrated tens of thousands of data sets across all these customers. And we have built an agent that enables our deployment strategy team to integrate this data agentically. And I think at this point, we have this amazing eval set where we can deterministically evaluate these agents for completeness and correctness on integrations. At this point, we're seeing about 90% of our, of our integrations. About 90% of that is written by agents with the oversight of our clients. of our deployment team. And these agents run for hours, right? And they will, you know, analyze, look at the databases, understand the ontology, start to piece together different pieces that need to be integrated. Those split off into sub agents that all do a bunch of work communicating back to kind of the orchestrator agent. Yeah. That's one bit on kind of long horizon agents. It reminds me of how people are using these coding agents for code-based migrations. Also, similarly, just... long-running, unglamorous work. 100%. I love this problem because it's verifiable. And that makes the problem a lot easier. This is why coding agents are, in a lot of ways, a lot easier. And... So the second piece that we spend a lot of time on is what do our agents look like for operational outcomes for our end users and the very first agent that we built was a cold case agent and uh This agent, you know, just give you some context on what some of these very high profile investigations look and feel like. You are uploading 200 to 300 gigabytes of data. These are videos, they are audio, it is images, it is a ton of PDF files and The detective is tasked to... with going through all that. So that in and of itself takes a really, really long time. I've been in these police departments where they actually have paper records of all these and they are boxes that are like this big. And, you know, they have all these CDs attached to it. It's really an outrageous amount of data. Mm hmm. And so our cold case agent, we first did, we built this with customer. And again, like this goes back to like how we like to how we like to build. And we were working with a customer who had worked a case where a man was wrongly convicted and they were able to exonerate. this individual. And they said, hey, can you reproduce this result with an agent? And so we took all of the evidence and data that they had on that case and started to work on an agent that would run for 30 minutes, 60 minutes to start to glean insights. And eventually we got to the place where it could reproduce the results that those detectives had gotten to. So that was where that was like kind of the the manifestation of our first kind of cold case agent. Now we're using this in a few departments across the US. We recently were working in a county in Wisconsin where This, again, similar type of case, 300 gigabytes of data. And they were able to identify and place the suspect at not just the scene of the crime, but where the body was found. And this was a single, there's about a few cell records, cell call detail records, kind of ping of a phone that was scattered amongst a lot of data. Thank you. points to this area where we could really up-level and help our public servants unburden them with all this administrative of watching hours and hours and hours of video and listening to hours and hours of audio. There's a quick comment. This example, Ben Brubb, is so impactful to the organization as a culture for our business. The way that we maintain trust is not actually taking credit and shouting from the rooftops about the awesomeness of what happened I think that is one of the fastest ways to break trust, frankly, with these organizations. The idea that we're going to scoop up the work and trumpet our skill and the way that we impacted the world. So I think being the quiet professionals in a context like this and empowering the customer is why we're going to be a part of this. [35:00] We have access to the next problem. I wish you would talk about it more, though, because I think that we're in this... moments where public distrust in AI is... So hi. Yeah. And... This is a wonderful story, right? I think that coming back to an earlier example of, um, uh, trying to identify threats to a synagogue. I think about... network effects. And I think about, um, What motivates us to tell our story? I... Um. I think the convergence of this idea of a kind of a centralized authoritarian AI capability and surveillance in American or national society, the convergence of these forces has created this immense pressure and I think distrust in many of the organizations and frankly, the communities that we work with. Mm hmm. that can't be ignored, first of all. And I think if we look at the the organizations we work with, and also the data landscape, the network effects are there and they are unbelievably strong. Like for example, it would be very easy for an organization that's trying to kind of hack their way to building something useful to go and hit open source, and pull in data that they might not, there might be a gray zone on whether they can or should access that based on laws, regulations, ordinances, standard operating procedures of, of the department. And for us, we, We have to protect against those moments where it's not like we don't need to break the rules, nor obviously should any organization break the rules in order to find a way to introduce AI or technology into these complex environments. The opposite is true. We have to protect the data, right? And we have to protect these organizations that by virtue of their structure have astounding proposition. Like you, how do you preserve the sanctity of the data and, and the ownership of the data and the ways of working for every individual. agency, every individual organization. And then building the connective tissue, building the interoperability is... going to happen. Hmm. And this is where things that might sometimes seem mundane, but we find really fascinating, like how do you think through data governance, how do you think through permissioning logic in the context of AI become unbelievably important? feel like, really motivated to talk about that in particular. Peregrine, the anti-network effects business. I think it's important, especially given the heat now around... AI is so powerful, and so... I love this data that used to exist. I think people now fear what happens when it's all... swooped up into a central panopticon that's deeply un-American. Yeah. Yeah. Yeah. Yeah. And I think that as a business, it's like I find that our business is a big practice in letting go. It's how do we build high integrity, transparent solutions for our customers? How do we make that transparent? parent. to our customers and their constituents. And then... How do we... effectively. let go and not try to grow too fast. Are there technology decisions that you have to make that are morally nuanced. So for example, Facial recognition. I'm curious your stance on that. And then more generally, what is your North Star for hard decisions? Quickly on facial recognition, I think the idea of a Silicon Valley company imposing a decision that is kind of general purpose for an industry is fundamentally wrong. I think the idea that we as an institution would assert, whether it be the utilization of a technology or a retention policy, I mean, it's very much not how we think. Thank you. to help a customer understand the context in which they're operating and actually bring to light all of the considerations they may or may not know and then help them. And this requires patience, help usher in the right way of doing business, the right way of thinking about deploying these technologies. On facial rec, most public safety agencies in America and their communities choose not to implement facial a matter of like subjective preference that they'll assert, uh, [40:00] But the idea of creating a technological red line without understanding the texture and context is not a boundary line that we think that we can assert on top of our customer. So follow the customer, follow the law. Yeah. And expose it and create clarity. around, you know, what left-right limits and, and what, what the institutional norms may be what i mean it's astounding how cities call each other they call each other for advice and so if you really just listen to the way that people learn inside of the industries that we serve what we realize is um Helping them streamline the way that they can get the best possible information so they can make their best decision, even about things like what technologies to use or not, goes a really long way. Maybe in a similar vein, I want to talk about delivering technology to the underdogs and how you're able to scale This very deep customer motion into a very different customer base than, you know, the other company that notoriously has scaled this motion Palantir, which notoriously doesn't take anything less than eight figure contracts. Mm hmm. Um, Did your mentors try to convince you not to do this? Did people tell you, like, the economics of this are not going to scale? And what made you think it was going to be possible to deliver technology to the underdogs and serve them in this way? I think it takes confidence. I think it takes confidence to, to believe that if you build something well, you can solve problems that have never been solved before that will scale later. And the idea of going deep, deep, to build a vertically integrated platform. tech stack that everything from network access to the permissioning and the governance and the ETL and the pipeline and the ontology and the UX and the APIs it's a to configure all of that and create a system that's open interoperable and can check these boxes it the audacity I think is is is is real um and And to build each of those components in a way that's first class, not just a check the box. I Truly thinking about how to deliver these types of technology solutions to an end market that has never been able to use these solutions before, let alone at the price point that we were able to deliver them. I think the uniqueness about technology. Peregrine is, for each organization, we'll take the time to deliver solutions. all the way to the outcome. And then we'll own the responsibility with our customers of getting to that outcome, which might seem extremely manual, extremely laborious, extremely unstable or unscalable. And... What we found is being able to do that resulted in a technology platform that could show up and do something very different for these organizations. and also scale. I've had the chance to talk to a few deployment strategists at Peregrine, and they're amazing. They're So smart, so humble. They so deeply embody the mission. And so I think you guys have selected for and trained a really extraordinary group of people. How do you do it? What do you look for? What do you interview for? How do you train these people up? First of all, I mean, it's the greatest compliment because I and we love our team. When you can build a culture... Where? people are trying really hard and can achieve excellence. It's almost like, how could you not create the environmental factors and... The feedback culture, which sometimes isn't easy, right? Sometimes it can be really intense. Ben and I would say, you know, we've aligned, we've taken tests and we've aligned on some character qualities, one of which is definitely the pursuit of excellence. I think a lot about these environmental factors of like trusting the individual contributor. It's a very... Um, important thing that we're trying to hold on to. And to do that with a forward deployed engineer, actually to truly do it really requires trust. Because we use this analogy of a dark cave, right? We'll send a person, into a cave and they have a tool belt. And hopefully the tools are pretty good. Right. And there might be a rope, you know, with a friend at the entrance of the cave that, you know, if they yell for help, you can kind of pull them out. But truly, we are sending people into these zones. And we're saying, Maintain your integrity. Like maintain your principles. Here are the tools. go do good and to do that at scale i really believe requires um Institutional trust and a belief system that innovation actually happens at the farthest fringes of our organization, that the way that our engineering product design organizations are fed is through that. innovation process across the country and now across multiple countries with our forward deploy team. [45:00] It's fascinating. So different from when I talk to companies who view Forward deploy this. at cost center. You guys are It's fundamentally very different talking to you. Definitely R&D. Definitely R&D and growth. I mean, after we land inside of a customer base, we have these models where we'll do these within customer pilots for all intents and purposes, where we'll keep sprinting on additional use cases. And in many ways, how we land and then how we expand in a customer base is fundamentally about leading from the front through our forward-deployed engineers. Have you discovered any new products through this motion? Mm-hmm. 100 I mean 100% it's just the whole thing like I think it's hard to even separate that I mean I remember walking or walking driving. across the Bay Bridge, San Pablo, typing on the computer and Nick's like, okay, we got to get these reports done. And this is what this detective needs. And literally typing as fast as I could to try and code what we needed at the time. And it wasn't theoretical. It was Aaron Blaisdell who had an extraordinarily urgent request. Yeah. And I feel it. And so it's so hard to separate. Very rarely are we in a room pontificating about what to build. I really think and how it impacts people. products, our product in two ways. One is... What are the primitives, basic primitives that the technology platform needs? And there's a couple of examples of this that I think are fascinating and reflect the ingenuity of our forward deploy team. For the longest time in Peregrine, we had no ability to edit. specific fields and properties. And so one forward deployed engineer to get around that made an integration. based on the comments people would write on these objects. And that comment would be read by Peregrine and then update a property. And that was the way they implemented editing. And so we saw that and we're like, we need to implement editing. And by the way, the reason that teammate did that is because... our intent was, and this is the way that the coaching and leading happens, it's like, your job is to hit the objective, like to hell with the technology. I mean, our job is to build the technologies and the tools that empower you to do higher and higher levels of work, but your job is to achieve. And that's what creates these crazy things that might look very unscaled, very hacky, but ultimately are the absolute best signaling network. It's great signal for what actually works. And it's like the quintessential rapid prototyping. So that's a category one. You get these these product primitives that you know that you need to provide the team in order to achieve the objective they're trying to achieve. I think like category two is... Almost... what I call it, innovation lab, where the deployment strategist goes and builds something totally unique that we're probably not going to integrate back into the product just because it's such a unique capability specifically for that customer. And this stuff gets me really excited. And I love watching and seeing what our deployment team builds. I was just looking at one The other day, where someone had built a hurricane simulator. in Peregrine. And I was like, how did you do that? I didn't know you could do that. And so that was really interesting. There was another deployment strategist who built this thing where you could It integrated a bunch of data from all these different sources, 911 call times, like budgets, and you could place a fire department in the city, and it would give you an estimate on how many people that would impact. It's unbelievable. And she was at the company for a month. Maybe last, last set of questions. 10 years from now, Suppose everything's gone right. Paragran is the institutional memory layer for 10,000 cities. I think that's great power that outlasts any one administration. And with great power comes great responsibility. How do you think about that? I think there's the outcome and then I think there's the path to get there. The end game of 10,000 cities... Um, Uh... I think requires an operating model that is fundamentally, um, about, uh, infrastructure, actually. It's like we as a technology organization are delivering technology infrastructure that empower these organizations to do with their data as they would like and as they are required to do. And every city, every jurisdiction, I deeply believe is unique. And I think the idea of... preserving that uniqueness is actually beautiful. Hmm. And, and, So I'm [50:00] excited about that. How do you make each, you know, the potpourri of cities, make each one more Awesome. And they're a unique way. That's cool. The path to get there requires... serious integrity, moral compass, core values, um, Not just in words, but in... um our ways of working, our actions, and that never goes away. It never goes away. The way that we think about the next... marginal customer that we support. Um, requires a level of hand-holding and delicacy that candidly, I don't think was possible even five years ago because the marginal cost of doing what we do. to drop it below a million bucks a year, I mean, it's radical, right? Coming back to the origin stories... in many ways, you know, it's supporting state, county and city level public safety was the There are a lineage of failed business units of major astounding organizations that tried to do it and failed because to deliver these solutions in a way that's tailored at a price point that these organizations can afford was never possible. And so to create the efficiency to drop the price by an order or orders of magnitude has earned us the right solution. to try to deliver on what you just said. It's like, who are we to think that we hold any power over that? I mean, it's the institution that has the power. Go make every city awesome. I think you're in a very, very serious seat. And... Just in the course of this conversation, it's clear that you're approaching this with with great care and great stewardship. So thank you for what you do. I think the world needs more examples of AI being used. to improve lives, to improve communities. And thank you for what you're doing. And thank you for joining the pod today. Thank you, Sonia. Music.