OpenCare

Health-aware AI chat

Every conversation is grounded in the patient's own data — demographics, medications, allergies, blood type, and conditions are injected into the Gemini context window automatically. The AI knows who it's talking to.

Streaming responses

Gemini's output is streamed token-by-token through the Express server using server-sent events, rendering progressively in the chat UI so patients see the response being written in real time rather than waiting for a complete answer.

Web search grounding

Optionally, the AI can call Tavily to fetch live web results and cite them alongside the patient's own health data — useful for questions about drug interactions, recent guidelines, or general medical context.

File attachments in chat

Patients can attach documents directly in the chat window. Files go through the full redaction → storage → embedding pipeline before the AI references their content in its reply.

Auto-titled conversations

After the first exchange, the server asks Gemini to generate a short descriptive title for the conversation and persists it — keeping the chat history organized without any patient effort.

PII redaction before storage

No sensitive term ever touches the object store unredacted. The Python microservice intercepts every upload, processes it with PyMuPDF to black out the specified terms, and only the sanitized version is forwarded to MinIO.

Semantic search via pgvector

Document chunks are embedded with a Hugging Face model and stored as vector columns in PostgreSQL using the pgvector extension. At query time, cosine similarity search retrieves the most relevant chunks to include in the AI's context.

S3-compatible file storage

MinIO provides S3-compatible object storage for all patient documents. Files are organized into per-user folders; downloads use pre-signed URLs so the frontend never needs direct storage credentials.

Structured health records

Patients maintain a profile covering date of birth, blood type, height, weight, current medications, known allergies, and chronic conditions. This structured data is always available to the AI alongside any uploaded documents.

Multi-provider authentication

Better Auth handles session management with support for email/password and Google OAuth. Multi-factor authentication is available. Patients can update their email and manage active sessions from the settings page.

AI preference settings

Patients can tune how the AI behaves — adjusting response style, verbosity, and whether web search grounding is active — from the settings page, without touching any configuration files.

Live database schema visualization

A dedicated /database route fetches the full live schema — every table, column, primary key, foreign key, and unique constraint — from the backend and renders it as an interactive entity-relationship diagram using Cytoscape.js.

Docker Compose orchestration

PostgreSQL, pgvector, MinIO, and all three application services are wired together in a single Compose file. The opencare-platform monorepo ties the submodules together so the entire system boots with one command.