Zoie
Zoie is a personal health intelligence platform that lets patients interact with their own medical records through a conversational AI interface.
Core trick
Ingests medical PDFs and notes into a vector store, then answers patient questions with multi-agent RAG over their own records.
Description
Zoie is a personal health intelligence platform that lets patients interact with their own medical records through a conversational AI interface.
Medical records like lab PDFs, radiology scans, doctor notes, prescriptions are scattered, hard to read, and clinically opaque to most patients. There's no easy way to ask "what does my LDL mean in context of my full history?" or "what did my last scan impression say?" Zoie bridges that gap with a multi-agent RAG platform that ingests a patient's documents, embeds them into a vector store, and lets them query their own health data in plain language.
README excerpt
# google-io-mediagent > [!IMPORTANT] > **New to the codebase?** Read the [**SETUP_GUIDE.md**](./SETUP_GUIDE.md) first. It is the "Bible" for getting this project running from scratch. A Domain-Driven Design (DDD) platform for managed agents: FastAPI + LangGraph backend, Next.js frontend, optional Telegram bot, and standalone research agents. ## Architecture | Component | Stack | Role | |-----------|--------|------| | **Backend** | FastAPI, LangGraph, pgvector | Agent registry, chat orchestration, ingestion, interpreter, check-ins | | **Frontend** | Next.js, Tailwind v4, shadcn/ui | Chat UI and agent management | | **Database** | PostgreSQL + pgvector | Health records, embeddings, demo data | | **Telegram** | python-telegram-bot | File understanding and API bridge (optional) | | **Research agent** | Gemini + public medical APIs | Standalone medical RAG prototype | Domain code lives under `backend/app/domains/` (for example `agent_registry`, `orchestration`, `ingestion`, `telegram`, `agents/research_agent`). --- ## Four-agent roadmap The platform now has four specialized agent packages under `backend/app/domains/agents/`. Research and deep insights are the fuller pipelines; scans and reports are intentionally hackathon-safe v1 agents that use simple text-only pgvector retrieval over patient records. | Agent | Status | Where it lives today | |-------|--------|----------------------| | **Research** | Packaged | `backend/app/domains/agents/research_agent/` — public APIs (