AI / RAG
DevMind AI — Research-to-Code Intelligence Platform
Transform research papers and GitHub repositories into actionable knowledge. Drag-drop a PDF, paste a GitHub URL, or write text — then chat with it via a streaming RAG pipeline powered by Groq LLaMA-3.
Public demo
4 — AI chat modes
The problem
Researchers and engineers spend hours manually extracting insights from dense papers or unfamiliar codebases. There's no fast path from 'raw document' to 'I understand this well enough to build on it.'
The approach
FastAPI backend ingests PDFs, GitHub READMEs, and raw text, chunks them semantically, and embeds each chunk with all-MiniLM-L6-v2 into a per-user FAISS index. At query time, top-K retrieval feeds context into Groq llama3-8b-8192 via SSE streaming. Four AI modes (Smart / Beginner / Deep Technical / Code Implementation) adjust the prompt persona. A comparison mode runs two documents side-by-side.
The results
End-to-end RAG with real-time streaming responses. JWT-secured multi-user auth. Side-by-side document comparison. Mobile-responsive glassmorphism UI. PostgreSQL in production, SQLite in dev — swapped via SQLAlchemy.
Stack: Next.js 14, TypeScript, FastAPI, FAISS, sentence-transformers, Groq API, SQLAlchemy