Knowledge Hub - AI-Powered Document Management System
A local-first document management system built with Flask, SQLAlchemy, and Postgres+pgvector. Supports OCR, semantic search, and LLM-powered Q&A for academic research.
How It Works
Project Overview
Knowledge Hub is a document management system I built to handle my MS in CS coursework at USC. The goal was simple: make it easy to search through course materials, research papers, and notes without digging through folders. It runs on Flask and PostgreSQL with the pgvector extension, so it supports both regular full-text search and vector-based semantic search. I added OCR processing for PDFs and images, document chunking, and hooked it up to a local LLM (Ollama) for question answering over uploaded docs. The whole thing is containerized with Docker. It has made finding relevant information across my coursework way faster than manual searching.
Document Upload & Processing
Key Features
Document Management
Upload, store, and organize documents with automatic metadata extraction and categorization
AI-Powered OCR
Automatic text extraction from PDFs and images using OpenCV, PyMuPDF, and Tesseract
Semantic Search
Vector-based similarity search using pgvector and Sentence-Transformers for intelligent content discovery
Question Answering
RAG-powered Q&A system with local LLM integration for contextual answers with citations
AI-Powered Question Answering
Technical Implementation
- Flask REST API with SQLAlchemy ORM for backend services
- PostgreSQL with pgvector extension for vector similarity search
- Docker containerization for easy deployment and scaling
- OCR processing with OpenCV, PyMuPDF, and Tesseract
- Vector embeddings using Sentence-Transformers (all-MiniLM-L6-v2)
- Local LLM integration with Ollama (gemma3:1b)
- Hybrid search combining full-text and semantic search
- Intelligent document chunking (300-700 tokens with overlaps)
- Confidence-aware ranking for OCR results
- RESTful API for document upload, search, and management
Challenges Faced
- Implementing efficient vector search with large document collections
- Optimizing OCR processing for various document types and quality
- Designing hybrid search algorithms for optimal relevance ranking
- Managing memory requirements for vector embeddings and LLM inference
- Creating intuitive API design for complex document operations
Key Learnings
- Advanced database design with vector extensions
- AI/ML integration in production web applications
- Document processing and OCR optimization techniques
- Vector database operations and similarity search algorithms
- RAG (Retrieval-Augmented Generation) system implementation
- Container orchestration and deployment strategies
Technical Documentation
Knowledge Hub Technical Deep Dive
Technical documentation covering architecture, implementation details, and design decisions.