DocMind

DocMind is an AI-powered document intelligence platform that supports conversational querying across 9+ file formats including PDF, DOCX, PPTX, CSV, XLSX, JSON, and SQLite, all in one place. It ships two retrieval engines: a Standard RAG pipeline built on LangChain LCEL with FAISS, MMR retrieval, adaptive top-k, and an optional cross-encoder reranker, and an Agentic RAG engine powered by LangGraph that follows a retrieve, grade, rewrite, and generate loop for complex multi-hop questions.

For computational queries over tables and spreadsheets, DocMind bypasses retrieval entirely and runs exact pandas expressions, eliminating hallucinated math. Additional features include structured metadata extraction, a two-column document comparator, and a built-in MCP server that exposes the RAG pipeline as tools directly inside Claude Desktop and Cursor. All generation runs on Google Vertex AI using Gemini 2.5 Flash, with embeddings via gemini-embedding-001. RAG quality is evaluated on every CI run using DeepEval measuring Answer Relevancy and Faithfulness. The app is containerized with Docker, deployed on AWS ECS Fargate behind an Application Load Balancer, and ships a full GitHub Actions CI/CD pipeline with rolling deployments.