Riley Deploys Multi-Agent AI to Tackle Pediatric Care Shortage

The Memory Revolution: How Riley's AI Infrastructure Creates Truly Personalized Parenting Support — Ai4

Ai4

With pediatrician shortages limiting clinical access, multi-agent memory systems can replace frantic online searches by forecasting developmental milestones across the entire family unit.

AI-assisted summary of The Memory Revolution: How Riley's AI Infrastructure Creates Truly Personalized Parenting Support — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published on YouTube by Ai4 on August 25, 2025, this talk features Amanda DeLuca, Co-Founder and CEO of Riley. According to the description, DeLuca presents how Riley uses a proprietary multi-agent memory system, fusion RAG, and knowledge graph architecture to overcome generic parenting advice and deliver hyper-personalized guidance at scale.

## TL;DR Drawing from her own experience with postpartum depression and late-night search panic, Amanda DeLuca addresses a widespread crisis where nearly half of US parents report overwhelming stress amidst declining pediatric access. To bridge the gap between locked clinical literature and individual family dynamics, she created Riley, an AI platform designed around four key pillars: a multi-agent memory system, a fusion RAG and knowledge graph, a proactive recommendation engine, and a multidisciplinary Clinical Advisory Board. By modeling the holistic family unit rather than an isolated individual, Riley forecasts upcoming milestones to replace anxiety with proactive, clinically vetted support.

## Atmosphere & style The presentation opens with candid, relatable humor regarding DeLuca's tight time constraints before quickly dipping into the raw, vulnerable reality of 3:00 AM postpartum panic. From there, the narrative transitions into a sharp, authoritative technical pitch, using vivid metaphors like a high-speed pachinko board to explain complex data decomposition. The overall delivery feels brisk, pragmatic, and deeply empathetic, balancing clinical seriousness with the genuine relief of a parent building solutions for other parents.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=4s) The inspiration for Riley stemmed from Amanda DeLuca's experience with postpartum depression and late-night panic over conflicting parenting advice. - [00:04](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=4s) Current online search results serve fragmented 'anecdata' and opinions rather than personalized, evidence-based science for parents. - [00:04](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=4s) 48% of parents in the US report feeling overwhelmed by daily stress levels, amid growing pediatrician shortages and social media misinformation. - [00:04](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=4s) Approximately 80% of questions submitted by parents to Riley are lightly medical in nature. - [02:45](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=165s) Riley combines taxpayer-funded clinical research and medical literature with granular household context such as work status, location, and children's exact ages. - [02:45](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=165s) A Clinical Advisory Board of pediatricians, neurologists, OB-GYNs, and therapists governs and vets all guidance generated by Riley. - [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) Riley's infrastructure utilizes a proprietary multi-agent memory system designed for advanced task decomposition. - [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) A fusion RAG and knowledge graph connects developmental science, family dynamics, and historical chat logs into an interconnected web. - [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) The platform incorporates a proactive recommendation engine that anticipates upcoming developmental leaps and transitions before parents search for them. - [07:53](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=473s) Unlike traditional AI systems built for single users, Riley is engineered to model and support the multi-variable holistic family unit.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=4s) The 3:00 AM Problem and the Crisis in Parenting Support

Amanda DeLuca introduces herself as the founder and CEO of Riley, an AI stage-based development app tailored for families. She opens on a lighthearted, urgent note—mentioning she has a strict 10-minute window because she needs to use the restroom right after—before diving into the personal origin story behind the product. Three years prior, DeLuca gave birth to her first child, a baby girl. Navigating the transition into parenthood while suffering from postpartum depression, she frequently experienced intense 3:00 AM panic moments: waking up startled by a baby's cry, an unusual sound, or recurring nightmares, only to fall down late-night Google rabbit holes frantically searching questions like, "Is she okay? Am I okay? Is she ever going to sleep? Is it okay that she's the 10th percentile?"

Searching for answers online proved to be an exercise in extreme frustration. The current internet landscape is not built for personalization, serving mostly fragmented "anecdata" and individual opinions rather than clear, evidence-based science. Existing parenting apps offered little relief, being overwhelmingly static, manual, poorly designed, and unintelligent.

This issue extends far beyond personal experience: - **48% of parents in the US** rate their daily stress levels as completely overwhelming, according to the Surgeon General. - Families are forced to navigate conflicting advice and misinformation from relatives and social media platforms like Instagram. - Access to high-quality pediatric care is declining both across the US and globally due to severe pediatrician shortages. - **Roughly 80% of queries** submitted to Riley are "lightly medical in nature," highlighting parents' reliance on tech for preliminary developmental and health concerns.

---

[02:45](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=165s) What Riley Is: Bridging Clinical Knowledge and Family Context

Recognizing that modern parents need more than generic search results or an isolated opinion, DeLuca set out to build an experience akin to having an on-demand panel of top experts. The goal was to pair evidence-based medical literature and developmental science—materials funded by taxpayer dollars yet often locked away from non-medical parents—with deep context regarding what makes an individual household unique.

```mermaid flowchart TD accTitle: Riley Context and Architecture Overview subgraph Inputs["Data Ingestion"] A["Family Context<br>(Work status, location, kids' ages/genders)"] B["Clinical Science & Literature<br>(Taxpayer-funded research & pediatrics)"] C["App Interactions & Chat History"] end

subgraph CoreEngine["Riley AI Core"] D["Multi-Agent Memory System<br>(Task Decomposition)"] E["Fusion RAG & Knowledge Graph<br>(Contextual Web)"] F["Proactive Recommendation Engine<br>(Stage Forecasting)"] end

subgraph Governance["Clinical Oversight"] G["Clinical Advisory Board<br>(Pediatricians, Neurologist, OB-GYNs, LCSWs)"] end

Inputs --> CoreEngine Governance -. "Vets & Guides".-> CoreEngine CoreEngine --> H["Hyper-Personalized Guidance<br>('Everything you need, nothing you don't')"] ```

Riley marries these two worlds using document, semantic, and visual connection-based memory. By capturing parameters such as parental working status, geographic location, household composition, and the exact ages and genders of the children, the platform filters out unnecessary noise to tell parents exactly what they need to know for their situation—and nothing they do not. The recommendations are strictly governed by a Clinical Advisory Board comprising pediatricians, a pediatric neurologist, lactation consultants, OB-GYNs, LCSWs, and PhDs.

---

[04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) The Four Architectural Pillars Powering Riley

Human parents do not evaluate a child's struggles in a silo; they connect sleep regressions to concurrent growth spurts, teething, or appetite shifts. Riley mimics this cognitive web of interconnected data points. Rather than merely knowing that an eight-month-old typically needs 14 hours of sleep, Riley integrates contextual knowledge—such as whether the infant recently cut a tooth or started pulling up to stand—to tailor its guidance.

```mermaid flowchart LR accTitle: Four Key Elements of Riley A["1. Multi-Agent Memory<br>Advanced task decomposition"] --> B["2. Fusion RAG & Graph<br>Unified knowledge web"] B --> C["3. Proactive Engine<br>Stage forecasting"] C --> D["4. Clinical Advisory Board<br>Trust and vetting"] ```

Riley's infrastructure relies on four core elements: 1. **Multi-Agent Memory System:** A proprietary architecture built from scratch (and rebuilt twice) designed for advanced task decomposition, functioning like a high-speed "pachinko board" that routes queries to fetch the exact context needed. 2. **Fusion RAG & Knowledge Graph:** A connected web linking evidence-based developmental milestones, historical chat transcripts, and idiosyncratic family dynamics into a single contextual layer. 3. **Proactive Recommendation Engine:** A stage-anticipation system that forecasts upcoming milestones and sleep or behavioral transitions before parents even search for them, helping replace anxiety with anticipation. 4. **Clinical Advisory Board Collaboration:** The medical backbone that verifies every recommendation to ensure guidance mirrors what a top pediatrician or specialist would advise.

---

[07:53](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=473s) The Future of Parenting Tech: Moving from Single Agents to Holistic Family Units

Addressing how Riley differentiates itself from generic AI tools and chatbots, DeLuca emphasizes its proactive anticipation and its distinct framing of the end user. While most AI agent architectures focus on assisting a single individual, Riley is engineered around the *holistic family unit* as a multi-variable system across thousands of households at scale.

DeLuca envisions a broader shift in the tech ecosystem toward empowering parents with high-fidelity, trusted data. In an era where direct access to pediatricians has become increasingly constrained compared to a decade ago, intelligent systems like Riley serve as an essential lifeline to foster confident, joyful, and well-supported parenting.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=4s) Is she okay? Am I okay? Is she ever going to sleep? Is it okay that she's the 10th percentile? — Amanda DeLuca

> [00:04](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=4s) The current internet landscape is not built for personalization, serving mostly fragmented 'anecdata' and individual opinions rather than clear, evidence-based science. — Amanda DeLuca

> [00:04](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=4s) 48% of parents in the US rate their daily stress levels as completely overwhelming — Amanda DeLuca

> [00:04](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=4s) Roughly 80% of queries submitted to Riley are lightly medical in nature — Amanda DeLuca

> [02:45](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=165s) Everything you need, nothing you don't — Amanda DeLuca

> [02:45](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=165s) Riley is powered by a proprietary multi-agent memory system that learns, adapts, and grows alongside each family. — Amanda DeLuca

> [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) functioning like a high-speed 'pachinko board' that routes queries to fetch the exact context needed — Amanda DeLuca

> [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) forecasts upcoming milestones and sleep or behavioral transitions before parents even search for them, helping replace anxiety with anticipation — Amanda DeLuca

> [07:53](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=473s) Riley is engineered around the holistic family unit as a multi-variable system across thousands of households at scale. — Amanda DeLuca

> [07:53](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=473s) transcends traditional data logging to deliver hyper-personalized guidance at scale, while respecting the irreplaceable human elements of parenting intuition and connection — Amanda DeLuca

## Mentioned Amanda DeLuca, Riley, Ai4, Google, Instagram, Surgeon General

## Video description Amanda DeLuca, Co-Founder, CEO, Riley

Abstracts

At 3AM with a crying baby, Amanda DeLuca discovered the fundamental flaw in parenting support: generic advice that ignores individual family contexts. This led her to build Riley, the first parenting app that can hyper-personalize at scale. Riley is powered by a proprietary multi-agent memory system that learns, adapts, and grows alongside each family. In this talk, Amanda reveals how Riley's cutting-edge fusion RAG retrieval, knowledge graph architecture, and proactive recommendation engine have revolutionized parenting support. She'll share how their technology transcends traditional data logging to deliver hyper-personalized guidance at scale, while respecting the irreplaceable human elements of parenting intuition and connection.

[Source](https://www.youtube.com/watch?v=0bd9p-XLrpE)

中文简报

Riley 部署多智能体 AI 应对儿科医疗资源短缺 在儿科医生短缺限制临床就医渠道的背景下，多智能体记忆系统可通过预测整个家庭单元的发育里程碑，取代令人慌乱的网络搜索。 结合自身经历产后抑郁及深夜搜索恐慌的切身体验，Amanda DeLuca 致力于应对一场普遍危机——在儿科就医机会日益减少的背景下，近半数美国父母表示承受着巨大的压力。为了弥合封闭的临床文献与个体家庭动态之间的鸿沟，她创立了 AI 平台 Riley。该平台围绕四大核心支柱构建：多智能体记忆系统、融合 RAG 与知识图谱、主动推荐引擎以及多学科临床咨询委员会。通过对整体家庭单元而非孤立个体进行建模，Riley 能够预测即将到来的成长里程碑，以主动且经过临床验证的支持来消除焦虑。 Amanda DeLuca 分享了创立 Riley 的初衷，源于其自身的产后抑郁及面对碎片化育儿建议时的焦虑；目前全美 48% 的父母承受着巨大压力，且向平台提出的咨询中有约 80% 属于轻度医疗性质。 (0:04) Riley 将公共循证临床医学文献与家庭的微观背景（如工作状态、地点及孩子的具体年龄）深度结合，并由儿科医生和神经学家等组成的临床顾问委员会对所有指导意见严格把关。 (2:45) 该应用的基础架构依托四大核心支柱：用于高级任务分解的专有多智能体记忆系统、融合 RAG 与知识图谱、主动式阶段预测推荐引擎，以及临床专家团队的严谨审核。 (4:32) 不同于仅针对单一个体的常规 AI 助手，Riley 将整个家庭单元视为多变量系统进行建模，旨在儿科医疗资源日益短缺的当下为家庭提供前瞻性且可信赖的支持。 (7:53)

Reading summary

Amanda DeLuca, founder and CEO of Riley, argues that modern parenting support is compromised by fragmented internet advice and pediatric care shortages, necessitating an AI infrastructure that bridges clinical literature with individual family context. The Information Crisis in Modern Parenting (0:04) DeLuca highlights a structural failure in traditional digital parenting resources, where late-night parental anxiety is met with static apps, unverified online opinions, and conflicting social media advice rather than actionable developmental science. According to data cited from the U.S. Surgeon General, 48% of parents in the United States describe their daily stress levels as completely overwhelming. Compounding this stress is a growing shortage of pediatricians both domestically and globally. With direct clinical access increasingly restricted, parents frequently turn to consumer technology for early triage; approximately 80% of user queries submitted to Riley are lightly medical in nature. Riley's Four-Pillar AI Architecture (4:32) To address these challenges, Riley connects taxpayer-funded developmental and clinical research with the specific context of individual households. The platform relies on a four-part architecture: a proprietary multi-agent memory system designed for task decomposition and query routing, and a Fusion RAG and knowledge graph that integrates historical chat transcripts, developmental milestones, and family dynamics. The system also features a proactive recommendation engine that forecasts developmental milestones and sleep transitions before users search for them. Medical oversight is embedded via a Clinical Advisory Board—including pediatricians, a pediatric neurologist, OB-GYNs, LCSWs, and PhDs—which verifies recommendations to ensure alignment with standard pediatric guidance. Modeling the Family as a System (7:53) DeLuca contrasts Riley with conventional AI chatbots by emphasizing its focus on the broader household. While standard agent architectures assist single individual users, Riley treats the family as an interconnected, multi-variable system across thousands of households. By tracking variables such as parental work status, geographic location, and the precise ages and genders of children, the platform links disparate signals—such as correlating sleep disruption with teething or physical milestones—to eliminate irrelevant noise and deliver tailored developmental guidance. DeLuca envisions parenting platforms shifting permanently from reactive search tools to proactive support systems, serving as necessary intermediaries in an era of strained pediatric infrastructure.

精炼阅读摘要

Riley 创始人兼首席执行官 Amanda DeLuca 指出，面对互联网育儿信息的碎片化以及儿科医疗资源的严重短缺，现代家庭承受着巨大的育儿焦虑。她主张通过结合多智能体记忆、融合检索增强生成（Fusion RAG）与临床医学验证，构建一个以家庭整体为核心单元的前瞻性 AI 支持系统。 深夜育儿焦虑与医疗资源的系统性短缺 (0:04) 根据美国卫生局局长的数据，48% 的美国父母认为其日常育儿压力水平已经达到“完全无法承受”的程度。与此同时，由于儿科医生的严重短缺，美国家庭乃至全球范围内获取高水平儿科医疗服务的难度不断上升。在面对婴儿发育指标、睡眠异常或突发哭闹等常见困扰时，家长往往被迫在社交平台与搜索引擎上盲目检索，陷入充满个别传闻、主观观点及矛盾信息的怪圈。 DeLuca 透露，向 Riley 平台提交的提问中约有 80% 在本质上“带有轻度医疗性质”，这凸显出家长在寻求初步健康与发育指导时对科技工具的高度依赖。然而，市面上现存的育儿应用普遍存在功能静态、高度依赖手动输入、设计不良且缺乏智能化交互等弊端，无法在父母深夜焦虑的时刻提供及时且具有循证医学依据的解答。 临床循证科学与家庭个性化情境的融合 (2:45) 为打破传统育儿搜索工具的局限，Riley 的设计理念是将原本由公共财政资助却常对普通人筑起门槛的临床医学文献与儿童发育科学，与各个家庭独特的微观情境深度绑定。该平台通过捕获父母的工作状态、地理位置、家庭成员结构以及孩子的具体年龄和性别，利用文档、语义和视觉关联记忆过滤无关噪声，力求做到“只呈现所需信息，剔除多余干扰”。 为了保证输出内容的权威性与安全性，Riley 建立了由临床顾问委员会主导的监督机制。该委员会成员涵盖儿科医生、儿科神经科医生、泌乳顾问、妇产科医生、持证临床社会工作者（LCSW）以及相关领域博士，全程指导并审核系统生成的建议，确保平台输出的指导与高水平专科医生的当面咨询意见保持一致。 四大技术支柱与互联认知模型 (4:32) 真实世界中父母对孩子的观察通常是网状互联的，例如会将睡眠倒退与正在经历的猛长期、出牙或食欲变化联系在一起。Riley 试图模拟这种人类认知网络，其底层架构依赖四大核心支柱：第一是自主研发并历经两次重构的多智能体记忆系统（Multi-Agent Memory System），该系统具备高阶任务分解能力，类似高速运转的“弹球盘”，能精确路由并提取所需的上下文背景；第二是融合检索增强生成与知识图谱（Fusion RAG & Knowledge Graph），将循证发育里程碑、历史对话记录与家庭个性化动态整合成统一的上下文网络。 第三项支柱是前瞻性推荐引擎（Proactive Recommendation Engine），该引擎能对即将到来的生长发育阶段、睡眠或行为模式转变进行预警预测，在父母主动检索前提前推送应对指引，将事后焦虑转化为事前准备；第四项支柱则是与临床顾问委员会的深度协作，构筑起严密的医疗风控背书，确保逻辑推理符合临床规范。 范式转变：从孤立单代理迈向多变量家庭单元 (7:53) 在谈及与通用 AI 工具及常见聊天机器人的本质区别时，DeLuca 强调 Riley 并不是针对孤立个体设计的单任务系统，而是将“整个家庭单元”建模为一个多变量的动态协同体系。目前绝大多数 AI 代理架构主要服务于单个用户的特定操作，而家庭级 AI 技术需要在数以千计的家庭规模上，综合考量家庭各成员间的相互影响与长期发展变数。 DeLuca 认为，面对直接获取儿科医疗资源日益受限的现实，科技生态必须向提供高保真、可信赖数据的方向发展。通过将家庭视为有机演化的整体并实现主动阶段预测，智能系统有望从被动应答突发提问的工具，转变为帮助父母建立信心、缓解养育压力的持续性支持网络。 DeLuca 期望通过深度整合临床证据与家庭上下文，使 Riley 成为缓解家庭育儿重压与儿科资源匮乏的关键纽带。然而，当平台承接高达 80% 的轻度医疗属性咨询时，如何在提供前瞻性个性化支持的同时严格厘清非医疗诊断的边界，并确保多智能体在复杂家庭情境下的决策精度，仍是此类家庭智能系统在大规模应用中需要持续验证的课题。

Detailed notes — zh

## Background 该内容由 Ai4 官方频道于 2025 年 8 月 25 日发布在 YouTube 平台。视频记录了 Riley 联合创始人兼 CEO 阿曼达·德卢卡（Amanda DeLuca）的主题演讲，简介介绍了 Riley 如何凭借多 Agent 记忆系统、Fusion RAG 与主动推荐引擎等 AI 架构，打破传统育儿建议千篇一律的缺陷，实现大规模的超个性化家庭育儿支持。 ## TL;DR 演讲中，Amanda 分享了自己产后遭遇育儿焦虑与信息过载的亲身经历，指出了当下近半数美国父母育儿压力过载及优质儿科医疗资源匮乏的困境。为此，团队打造了以家庭为核心单元的 AI 育儿应用 Riley，依托自主开发的多 Agent 记忆系统、Fusion RAG、知识图谱以及跨学科临床顾问委员会，综合分析家庭的多维上下文变量。不同于市面上被动响应的单人 AI 工具，Riley 能够基于发育规律前瞻性预测成长里程碑并主动介入，致力于用科学个性化的智能支持赋能新一代充满自信的父母。 ## Atmosphere & style 整场演讲节奏明快且极具共情力，Amanda 以“凌晨三点新手父母陷入搜索引擎恐慌”的真实痛点幽默开场，迅速引发全场共鸣。随着技术拆解与临床把关机制的深入，分享自然转向严谨、干练且充满自信的科技路演风格。生动的情感切入与清晰的技术愿景紧密交织，既传递出新手父母面临的脆弱与真实压力，又展现了家庭级 AI 架构为现代育儿带来的笃定感与温暖支持。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=4s) 传统互联网育儿搜索充斥着主观片面的经验（anecdata）与无效焦虑，市面现有工具机械落后，促使 Riley 探索高保真个性化育儿支持。 - [02:07](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=127s) 全美高达 48% 的父母承受着过载的育儿压力，且 Riley 平台上约 80% 的查询涉及轻度医疗健康咨询，反映出专业儿科资源的普遍匮乏。 - [02:07](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=127s) Riley 将家庭视作有机整体，将父母工作、居住环境、子女结构等全景家庭变量与儿童发展科学深度整合，提供个性化决策情境。 - [02:07](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=127s) 组建涵盖儿科神经学、妇产科、泌乳顾问及临床社工等跨学科专家委员会，运用语义与视觉连接记忆严格把关医学建议的权威性。 - [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) 采用仿生联想机制模拟人脑思维，将单一行为异常（如夜醒、睡眠倒退）与生长突增、出牙等动态网络关联，避免机械僵化的教条判断。 - [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) 自研多 Agent 记忆系统，通过类似“弹珠机”的任务分解与路由机制，高效解构复杂的家庭情境并调度精准数据。 - [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) 依托 Fusion RAG 与知识图谱技术，将用户日常交互历史、对话上下文与权威医学知识库深度交织为情境化洞察网络。 - [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) 主动推荐引擎利用群体数据沉淀与发育规律，在父母提问之前预判发育里程碑与阶段挑战，实现由被动解答向主动介入的跃迁。 - [07:53](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=473s) 超越主流 AI 的单人问答模式，Riley 专注构建家庭级预测架构，有效填补母婴育儿科技赛道中优质智能化产品的长期缺位。 - [09:01](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=541s) 终极愿景在于借助深度的个性化数据后盾，培养充满力量与自信的新一代父母，从容应对儿科医疗资源紧缺等现实困境。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=4s) 凌晨三点的育儿焦虑与 Riley 的诞生初衷 在第二次登上 AI4 演讲台时，Riley 联合创始人兼 CEO 阿曼达·德卢卡（Amanda DeLuca）带着幽默与快节奏开场。她坦言接下来的分享必须精简高效，因为留给她的演讲时间只有 10 分钟。Amanda 创立的 Riley 是一款面向家庭的 AI 阶段化儿童发展应用，其核心壁垒在于团队从零自主构建的专属记忆客户端（bespoke memory client），这项技术让高保真、大规模的个性化育儿支持成为可能。 谈及创业灵感，Amanda 分享了三年前诞下大女儿时的亲身经历。从非父母状态转变为父母的剧烈转变让所有人感到窒息，加之产后抑郁（postpartum depression）的折磨，她经常在凌晨三点遭遇育儿噩梦。每当女儿发出奇怪的声响或哭闹，她便从惊恐中猛然坐起，陷入搜索引擎的“末日信息流”（doom scroll），疯狂输入诸如“孩子正常吗？我正常吗？她到底能不能睡整觉？体重在第 10 百分位数（10th percentile）有危险吗？”等问题。 她很快意识到，在早期儿童成长和育儿领域，现有互联网根本无法提供精准解答。网上海量充斥着个人的主观意见与片面经验（anecdata），家长极难辨别哪些是真正有科学依据的核心事实。与此同时，市面上现存的育儿工具普遍静态、机械、缺乏智能，且界面与架构设计落后，这让新手父母在最脆弱的时刻陷入了极度沮丧。 --- ### [02:07](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=127s) 全美家长的系统性困境与家庭级上下文感知 这种焦虑并非个例。根据美国卫生局局长（Surgeon General）的数据，全美有高达 48% 的父母将日常育儿压力评为“完全无法承受（completely overwhelming）”。许多家庭向 Riley 反馈，他们每天都在应对来自长辈以及 Instagram 等社交媒体的育儿错误信息；与此同时，全球范围内高水平儿科医疗资源的短缺日益严重。从 Riley 收集到的实际数据来看，约有 80% 的用户咨询带有轻微的医疗咨询性质，家庭迫切需要专业可信的指引。 基于自身的痛点，Amanda 意识到家长真正需要的不是单一专家的偶尔点拨，而是一个常驻身边的“专家评审团”：他们不仅精通顶尖医学文献与前沿科学，更深刻理解每个家庭的独特性。Riley 正是围绕这一理念构建——它不仅收录普通非医学人士难以触及的高门槛儿童发育科学，还将家庭的全部现实变量纳入计算，包括父母的工作状态、居住地点、子女数量、年龄性别乃至同住家庭成员等。 为了确保建议的高标准输出，Riley 建立了跨学科的临床顾问委员会（Clinical Advisory Board），团队内部包括全职儿科神经学家、儿科医生、泌乳顾问、妇产科医生、临床社工（LCSW）以及发展心理学博士。平台利用文档语义（document semantic）与视觉连接记忆技术，严格剔除冗余干扰，仅向家长输出当前情境下“真正需要知道的关键信息”。 - **48%** 的美国父母表示日常育儿压力处于完全过载状态。 - **80%** 的用户输入查询与轻度医疗及健康建议相关。 - 专家顾问团涵盖儿科、儿科神经学、妇产科、泌乳咨询及临床社工等多个专业领域。 --- ### [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) 仿生联想与四项核心技术支柱 人类大脑在思考育儿问题时从不孤立看待单一事件，而是将其置于动态的联系网络中。例如当孩子出现睡眠倒退时，父母会联想到这是否与出牙（teething）、食欲变化或生长突增期（growth spurt）有关。Riley 的记忆架构正是模拟这种思维方式：它不会机械地判定“一个八个月大的婴儿每天需要睡 14 个小时”，而是结合上下文明确知晓“这个八个月大的宝宝最近刚长了一颗牙，并且正在练习站立”，从而提供高度贴合个体现状的解决方案，让父母摆脱无谓的内耗。 这一机制由四大核心技术要素共同驱动： 1. **多 Agent 记忆系统（Multi-Agent Memory System）**：团队完全从零自主迭代重构了两次该系统，它展现了极强的高级任务分解能力，运作流程如同精密的“弹珠机（pachinko board）”，能高保真地快速判定当前家庭情境并调度必要的数据源。 2. **Fusion RAG 与知识图谱（Knowledge Graph）**：将用户交互历史、聊天上下文与权威医学知识库紧密交织，构建出一张可视化、情境化的洞察之网。 3. **主动推荐引擎（Proactive Recommendation Engine）**：基于大规模用户数据沉淀与儿童发展的规律性，在父母开口提问之前预测即将到来的发展里程碑与阶段挑战，化被动解答为主动引导。 4. **临床顾问深度协作**：Amanda 将自己定位为“Riley 的 0 号用户”，所有算法输出与内容规则均经过临床专家的严格把关，确保每条建议都经得起儿科医生的推敲。 ```mermaid flowchart TD accTitle: Riley AI Technical Architecture subgraph Inputs["输入与上下文层"] F["家庭全景变量<br/>(成员/职业/年龄/居住地)"] K["权威临床知识库<br/>(医学文献/发育里程碑)"] U["动态交互历史<br/>(对话记录/即时状态)"] end subgraph Core["核心处理引擎"] M["多 Agent 记忆系统<br/>(高级任务分解/弹珠机路由)"] R["Fusion RAG 与知识图谱<br/>(语义与情境关联网络)"] P["主动推荐引擎<br/>(发育阶段前瞻预测)"] end subgraph Assurance["专业审核与交付"] C["临床顾问委员会把关<br/>(儿科神经/妇产/泌乳专家)"] A["个性化精准育儿指引"] end F --> M U --> M K --> R M --> R R --> P P --> C C --> A ``` --- ### [07:53](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=473s) 超越单人 Agent：以家庭为核心的预测型育儿科技 当被问及 Riley 与市面上其他主流 AI 工具的本质区别时，Amanda 指出，绝大多数 AI 仅局限于“单人交互”和“被动问答”模式。而 Riley 是首个将“家庭”作为一个完整有机单元来处理的 AI 系统。它能够跨越成千上万个家庭，深入拆解每个家庭单元内部错综复杂的多变量关系。 更重要的是，Riley 具备前瞻预测能力，能够在问题发生前主动介入，在恰当的时间节点为家长提供预警与心理准备，帮助父母重新找回育儿中的从容与乐趣。Amanda 强调，当前的母婴育儿科技赛道仍处于优质产品匮乏的阶段，Riley 正在通过深耕家庭级记忆与预测架构，推动下一代家庭科技生态的建立。 ### [09:01](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=541s) 赋能新一代父母与 Riley 的愿景总结 Amanda DeLuca 总结指出，在 Riley 出现之前，市面上针对家庭场景的科技支持一直相当匮乏。她创立 Riley 的核心愿景，是希望能见证并培养出“一代极其充满力量、无比自信的父母”（a generation of incredibly empowered, incredibly confident parents）。通过让家庭获得这些深度的个性化数据与全方位的智能支持，父母们能够拥有坚实可靠的育儿后盾。 尤其是在当前优质儿科医疗资源与服务获取难度相比十年前更为严峻的背景下，这种智能支持体系显得尤为关键。即便面对外部儿科医疗资源的局限，父母们依然能够依靠 Riley 获得科学、及时的个性化指导。Amanda 最后以“这就是 Riley 的概貌”（that's Riley in a nutshell）为演讲画上句号，向全场观众致谢并在掌声中结束了分享。 - **核心愿景**：打造真正适用于家庭场景的 AI 技术，填补优质儿科护理资源获取不易的鸿沟，让父母更有底气与信心地应对育儿挑战。 - **价值支撑**：依托丰富的数据沉淀与个性化辅助体系，为家庭提供随时随地可信赖的决策支持。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=4s) 孩子正常吗？我正常吗？她到底能不能睡整觉？体重在第 10 百分位数（10th percentile）有危险吗？ — Amanda DeLuca > [02:07](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=127s) 完全无法承受（completely overwhelming） — Surgeon General > [02:07](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=127s) 真正需要知道的关键信息 — Amanda DeLuca > [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) 一个八个月大的婴儿每天需要睡 14 个小时 — Amanda DeLuca > [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) 这个八个月大的宝宝最近刚长了一颗牙，并且正在练习站立 — Amanda DeLuca > [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) 弹珠机（pachinko board） — Amanda DeLuca > [04:32](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=272s) Riley 的 0 号用户 — Amanda DeLuca > [09:01](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=541s) 一代极其充满力量、无比自信的父母（a generation of incredibly empowered, incredibly confident parents） — Amanda DeLuca > [09:01](https://www.youtube.com/watch?v=0bd9p-XLrpE&t=541s) 这就是 Riley 的概貌（that's Riley in a nutshell） — Amanda DeLuca ## Mentioned Amanda DeLuca, Riley, Ai4, Surgeon General, Instagram ## Video description Amanda DeLuca, Co-Founder, CEO, Riley Abstracts At 3AM with a crying baby, Amanda DeLuca discovered the fundamental flaw in parenting support: generic advice that ignores individual family contexts. This led her to build Riley, the first parenting app that can hyper-personalize at scale. Riley is powered by a proprietary multi-agent memory system that learns, adapts, and grows alongside each family. In this talk, Amanda reveals how Riley's cutting-edge fusion RAG retrieval, knowledge graph architecture, and proactive recommendation engine have revolutionized parenting support. She'll share how their technology transcends traditional data logging to deliver hyper-personalized guidance at scale, while respecting the irreplaceable human elements of parenting intuition and connection. [Source](https://www.youtube.com/watch?v=0bd9p-XLrpE)

Source quotes

[0:04] Amanda DeLuca: 孩子正常吗？我正常吗？她到底能不能睡整觉？体重在第 10 百分位数（10th percentile）有危险吗？ [2:07] Surgeon General: 完全无法承受（completely overwhelming） [2:07] Amanda DeLuca: 真正需要知道的关键信息 [4:32] Amanda DeLuca: 一个八个月大的婴儿每天需要睡 14 个小时 [4:32] Amanda DeLuca: 这个八个月大的宝宝最近刚长了一颗牙，并且正在练习站立 [4:32] Amanda DeLuca: 弹珠机（pachinko board） [4:32] Amanda DeLuca: Riley 的 0 号用户 [9:01] Amanda DeLuca: 一代极其充满力量、无比自信的父母（a generation of incredibly empowered, incredibly confident parents） [9:01] Amanda DeLuca: 这就是 Riley 的概貌（that's Riley in a nutshell）

Source quotes

[0:04] Amanda DeLuca: Is she okay? Am I okay? Is she ever going to sleep? Is it okay that she's the 10th percentile? [0:04] Amanda DeLuca: The current internet landscape is not built for personalization, serving mostly fragmented 'anecdata' and individual opinions rather than clear, evidence-based science. [0:04] Amanda DeLuca: 48% of parents in the US rate their daily stress levels as completely overwhelming [0:04] Amanda DeLuca: Roughly 80% of queries submitted to Riley are lightly medical in nature [2:45] Amanda DeLuca: Everything you need, nothing you don't [2:45] Amanda DeLuca: Riley is powered by a proprietary multi-agent memory system that learns, adapts, and grows alongside each family. [4:32] Amanda DeLuca: functioning like a high-speed 'pachinko board' that routes queries to fetch the exact context needed [4:32] Amanda DeLuca: forecasts upcoming milestones and sleep or behavioral transitions before parents even search for them, helping replace anxiety with anticipation [7:53] Amanda DeLuca: Riley is engineered around the holistic family unit as a multi-variable system across thousands of households at scale. [7:53] Amanda DeLuca: transcends traditional data logging to deliver hyper-personalized guidance at scale, while respecting the irreplaceable human elements of parenting intuition and connection

Original transcript

[0:04] Hello. Hello. I am so excited to be here [0:09] speaking at AI4 for the second time. Um, [0:11] and I'm also especially excited that [0:13] this chat is going to be 10 minutes cuz [0:14] I have to pee in like 10 minutes. So, [0:16] like, let's get this going. Um, I'm [0:18] Amanda Duca. I am the founder and CEO of [0:21] Riley, which is an AI stagebased [0:23] development app for families. And I'm so [0:26] excited, especially today, to talk about [0:28] a couple of the things that make Riley [0:30] incredibly unique, particularly the kind [0:32] of bespoke memory client that we've [0:34] built ourselves from scratch that [0:35] enables us to do incredible highquality [0:38] personalization at scale. So, let's get [0:40] into it. [0:42] So, just about exactly three years ago, [0:44] uh I had my first baby, a little girl. [0:46] And as any parents in the room know, um [0:49] the transition from not being a parent [0:51] to being a parent is incredible and [0:53] incredibly intense. And on top of all of [0:55] that, I had postpartum depression. And I [0:58] would find myself having these moments [1:00] that I think again are very familiar to [1:01] parents in the room um at 3:00 a.m. [1:03] where like maybe my daughter was crying [1:05] or she'd made a weird noise or I had had [1:07] one of my daily panic nightmares and I [1:09] just kind of bolted up right in bed and [1:11] would go down these Google like doom [1:13] scroll rabbit holes with questions like, [1:15] "Is she okay? Am I okay? Is she ever [1:19] going to sleep? Is it okay that she's [1:21] the 10th percentile?" And I really [1:23] quickly found that trying to find any [1:25] sort of good answers, especially about [1:27] kids, especially about like what really [1:29] does and doesn't matter in early [1:30] childhood and parenting, um was an [1:32] exercise in extreme frustration. The [1:34] internet as it exists today is really [1:36] not built for personalization. And also [1:38] I was like stumbling upon these articles [1:40] that were like one person's opinion or [1:43] anecdata. And I couldn't get a good [1:45] sense of like all right, what does the [1:46] data actually tell us? Like what do we [1:48] know makes a difference? Like what do I [1:50] need to know? And on top of all of that, [1:53] I would, you know, kind of go to these [1:55] apps, this existing tech for parents [1:56] that was just incredibly static and [1:58] manual and just not intelligent, usually [2:00] badly designed, badly built or both. And [2:02] so really kind of got into this very [2:05] frustrated state. And what's amazing [2:07] about this is that this is definitely [2:09] not an Amanda specific problem. In fact, [2:11] 48% of parents in the US today rate [2:14] their daily stress level as completely [2:16] overwhelming according to the surgeon [2:17] general. And Riley families tell us all [2:20] the time that, you know, they are [2:21] juggling misinformation from like family [2:23] members and Instagram, god forbid. And [2:27] that access to really high quality [2:28] pediatric care is declining overall just [2:30] in the United States and across the the [2:32] world. Um, there just aren't enough [2:33] pediatricians. Anecdotally, also, we see [2:36] that about 80% of queries that come into [2:38] Riley are lightly medical in nature. So, [2:40] parents are having a tough time here. I [2:42] certainly was when my daughter was born. [2:45] And this is where Riley comes in. And I [2:48] really what happened here was I realized [2:51] that what I wanted was a panel of [2:53] experts, not just one person. I wanted a [2:54] panel of experts who knew what the [2:57] science said, who had access to this [2:58] incredible highquality like [3:00] best-in-class medical literature, data, [3:01] and science, but who also knew me and [3:04] everything that made my family unique [3:05] and could tell me exactly what I needed [3:08] to know, nothing that I didn't need to [3:09] know, so I can move on with my life. And [3:11] that's exactly what Riley is. So Riley [3:13] takes everything that makes each [3:15] individual family unit unique. Things [3:16] like working status of parents, where [3:18] you live, how many kids you have, the [3:20] ages and genders of your kids, um who [3:23] lives with you. Uh as well as again this [3:26] amazing knowledge base of just that, you [3:29] know, what really confidently does and [3:31] does not matter in early childhood [3:33] development and parenting. Um a lot of [3:35] stuff that's really difficult for a [3:36] non-medical person to access even though [3:38] our tax dollars pay for them and [3:40] creates, you know, magic. at scale by [3:42] marrying those two things. So, what [3:45] we're able to do is use document [3:47] semantic and visual connection-based [3:50] memory to really tell you, the parent [3:52] and the family, like here's everything [3:54] that you specifically need to know for [3:56] your kids and your situation and nothing [3:59] that you don't. Also really amazing here [4:02] is that all of this is underpinned by [4:04] our incredible collaboration with our [4:05] clinical advisory board. you know, folks [4:07] who are absolute experts in the field, [4:08] pediatricians, we've got a pediatric [4:10] neurologist on staff. Um, we have [4:12] lactation consultants, obgynes, um, [4:15] LCSW, PhDs who really know again what [4:18] does the data say, what would they say [4:20] to patients coming in with questions or [4:22] needing specific types of support. So, [4:24] we have a very, very high bar for what [4:26] actually goes out in the Riley app and [4:28] what actually ends up coming to parents. [4:32] So think about the way that your brain [4:35] thinks about supporting your kid or [4:36] thinking about your family. You're not [4:38] just thinking about each individual [4:39] child or each individual instance in [4:41] like a silo in a vacuum. You're [4:42] thinking, "All right, we're going [4:44] through a sleep aggression and maybe [4:45] it's because she's teething or having an [4:48] appetite change or, you know, going [4:49] through a growth spurt." You're thinking [4:51] about all of these kind of inter [4:53] interconnected facts like as a really [4:56] nice web of like things that could be [4:58] related. And Riley's memory works very [5:00] very similarly to say kind of [5:02] holistically, all right, not just an [5:05] eight-month-old is probably getting 14 [5:06] hours of sleep. It knows that your [5:08] eight-month-old recently cut a tooth and [5:10] is maybe practicing standing. And so can [5:12] can, you know, again, very particularly [5:15] customized and personalized to your [5:17] specific case, your specific kids and [5:20] your specific family to really tell you [5:22] what you need to know so that you can [5:23] again go on about your day and spend [5:26] more joyful time with your kids. And [5:28] this is really powered by four key [5:30] elements. The first one is our amazing [5:33] multi- aent memory system. This is like [5:35] top of the top best-in-class. We built [5:36] it all ourselves from the ground up [5:38] twice. Um we also have a really cool [5:40] fusion rag system that we use to again [5:42] kind of take these you know things that [5:44] may be important for you to know things [5:46] from your you know interactions with the [5:48] app chat history as well as again that [5:50] kind of incredible knowledge base that [5:52] we have and put them together at scale. [5:54] We also have a proactive recommendation [5:56] engine that understands, you know, what [5:58] are the developmental stages that your [5:59] kids are going through and that are [6:00] coming up for them. So that as a parent, [6:03] you have a really good sense of [6:04] grounding for like what's currently [6:05] going on and also know what to expect. [6:08] The developmental stages that kids go [6:09] through, even though each kid is unique, [6:11] they tend to be fairly predictable early [6:12] on. And so we can give you a sense of, [6:14] hey, what's coming up for you? As well [6:16] as like things to be excited about [6:18] because a lot of parenting can be really [6:19] exciting if it's not too overwhelming. [6:22] And then finally we have our expert [6:24] collaboration again that clinical [6:26] advisory board who really helps us know [6:28] what is going to be a great quality [6:30] piece of guidance piece of advice. [6:33] Double clicking here the multi-agent [6:35] memory system is really an exercise and [6:37] like a a an example of a master class [6:39] and advanced task decomposition. I like [6:41] to think of it as like a pachinko board [6:42] where you know we're able to really [6:44] really kind of quickly and and and with [6:46] high fidelity say what are the things [6:48] that we need to know? What is what's [6:50] your situation? What do you need right [6:51] now? and what do we need to pull from [6:53] where in order to give you the best [6:54] possible advice. Our fusion rag and [6:57] knowledge graph again is that kind of [6:59] beautiful visual um connected web of [7:03] insight from our knowledge base from the [7:04] developmental stages as well as from [7:06] your individual family so that we can [7:08] say all right holistically contextually [7:10] what is this what do we need to know [7:12] what do you need to know and what do you [7:14] not need to know the proactive [7:17] recommendation engine again is just you [7:18] know understanding what's coming before [7:20] you even have to ask we can anticipate a [7:22] lot of this especially since we have so [7:24] much data now at scale from our users [7:26] And finally, our clinical advisory board [7:28] is really the piece that makes this all [7:30] come together. As user zero O'Reilly, I [7:33] want to be able to trust that the advice [7:34] that I'm getting, the recommendations, [7:36] the support that I'm getting is good. It [7:38] is trustworthy. It does align with what [7:40] my pediatrician might say, what with [7:42] what experts are saying about how to [7:44] handle each individual thing that [7:46] happens with my child. And so that [7:48] really is, you know, the support that [7:50] powers everything else. [7:53] And so we think about what's next for [7:54] parenting tech and what you know I'm [7:57] excited to see in just kind of the space [8:00] in general is you know a lot of times [8:02] people will ask me like hey how is this [8:04] different from any other AI tool that [8:07] exists and I love that question because [8:10] what Riley can do that no AI other AI [8:13] tool will do for you is it can again [8:15] anticipate and provide that proactive [8:17] support. It can kind of meet you where [8:19] you are and let you know what's coming [8:20] by giving you really deep insight that [8:22] no other tool could possibly have [8:25] because no other tool is, you know, [8:26] really built by and for families. And so [8:29] we also often talk about, you know, with [8:30] agents providing really individual [8:33] information to one person. Riley is [8:36] really the first system that I've seen [8:37] that thinks about a holistic unit, a [8:39] family, and the variables that make each [8:42] individual instance of that at scale [8:44] across thousands and thousands of [8:45] people, thousands of families unique and [8:48] worth again kind of decomposing in the [8:50] way that we do. And so I'm excited to [8:52] see more of that as well. And I'm also [8:54] just excited to see more interesting, [8:56] really good quality parenting tech in [8:57] general. We're still in a place where [8:59] there isn't great tech for parents other [9:01] than Riley. There isn't great tech for [9:03] families. And I want to be able to kind [9:06] of see a generation of incredibly [9:08] empowered, incredibly confident parents [9:10] who have access to this incredible data, [9:12] this incredible support, um, and, you [9:14] know, can kind of rely on that even if [9:16] their access to highquality pediatric [9:18] care is not, you know, where it would [9:20] have been 10 years ago. So that's Riley [9:23] in a nutshell. Very, very happy to be [9:26] here and thank you so much [9:29] [Applause]