DocuForge : Document Parser & Vector Search
Automated document ingestion, OCR parsing, and vector semantic retrieval for unstructured technical literature.
Overview
DocuForge simplifies technical document discovery by extracting text, tabular data, and metadata from PDF files, converting them into vector chunks, and serving a low-latency semantic search interface.
My role
AI & Full-Stack Engineer — Built the FastAPI extraction microservice, configured ChromaDB vector collections, and created the React search dashboard.
Key features
Context-Preserving Chunking
Recursive character chunking with sliding window overlap to maintain sentence integrity and semantic context across paragraph boundaries.
ChromaDB Dense Vector Search
Cosine similarity search over vector collections enabling instant relevance ranking and cross-document concept queries.
Async FastAPI Processing
Non-blocking background ingestion pipelines for large multi-page PDFs with real-time status webhooks.
Results
Enabled sub-second semantic retrieval across 500+ technical PDF documents with accurate source page and paragraph citations.