DocuMind

The Challenge

Organizations handle large volumes of scattered documentation in multiple formats and locations. Finding relevant information requires hours of manual searching. DocuMind solves this by providing semantic search that understands meaning, not just keywords, and generates contextualized responses citing original sources.

My Role

I lead the complete design and development of the system, which includes:

• Hybrid Architecture: Designing a local-cloud architecture that combines local LLM models (LM Studio) with cloud vector databases (Qdrant), offering a scalable and efficient solution for corporate knowledge management.
• RAG Pipeline: Implementing a complete RAG pipeline with LangChain that includes document ingestion, semantic chunking, embedding generation, and response generation with source traceability.
• Multi-format Processing: Developing processors for 25+ document formats including Office, PDF, OpenDocument, emails, and images with OCR using Tesseract for scanned documents.
• Real-time Monitoring: Implementing file watchers with automatic detection of new documents, intelligent deduplication, and support for multiple folders and network paths.
• Interactive Interface: Creating a complete dashboard with Streamlit that includes semantic search, document navigation, system configuration, and real-time statistics.

Seeking Investment

The project is in active development phase seeking investors to scale the solution. The modular and containerized architecture allows adaptation to different enterprise needs while maintaining quality and traceability.

Technical Architecture

The system uses Docker Compose to orchestrate multiple services: Qdrant for vectors, FastAPI for REST API, Streamlit for UI, n8n for workflow automation, PostgreSQL for persistence, and Redis for cache. Local LLM models run on LM Studio for privacy and control.