MindLens
The Vision
A privacy-conscious multi-agent wellbeing platform with bounded LangGraph orchestration, deterministic crisis protection, and hosted retrieval.
MindLens is a full-stack AI system for reflective mental-health support. I built the workflow around explicit nodes and conditional edges so safety, assessment, memory recall, retrieval, specialist generation, review, and persistence remain inspectable and bounded. The hosted version combines FastAPI, Next.js, Groq, Pinecone, MongoDB Atlas, and metadata-only Langfuse tracing.
What I Built
- 01
Deterministic crisis screening exits before any model or retrieval call.
- 02
PII is anonymized before external model calls; personal conversations never enter Pinecone.
- 03
Normal turns are bounded to a maximum of three Groq calls, with graceful degradation when providers fail.
- 04
Stable REST and WebSocket contracts connect the FastAPI backend to the Next.js frontend.