AI-Powered Document Assistant

RAG-based document Q&A. Upload a PDF, ask questions about it, get answers grounded in the document with page-level citations.

Architecture

Upload -> PDFBox text extraction (page-aware) -> chunking (overlap) -> embeddings (Ollama, nomic-embed-text) -> Postgres/pgvector storage -> retrieval (cosine similarity, HNSW index) -> LLM answer generation (Ollama, qwen3:8b) with page citations.

Stack

• Spring Boot 3 / Java 17
• Plain JDBC (JdbcTemplate), no JPA - pgvector's similarity operators aren't something JPQL/Hibernate understands natively
• Postgres + pgvector for storage and similarity search
• Apache PDFBox for page-aware text extraction
• Ollama, fully local and free - nomic-embed-text (768-dim) for embeddings, qwen3:8b for answer generation. No API key, no external dependency.

Endpoints

• POST /api/documents - upload a PDF, runs the full ingest pipeline
• POST /api/documents/{id}/retrieve - top-5 most similar chunks for a question (debug endpoint, no LLM call)
• POST /api/documents/{id}/ask - full RAG: retrieve + generate + cite

Run locally

1. docker compose up -d
2. ollama pull nomic-embed-text (qwen3:8b assumed already installed)
3. mvn spring-boot:run

Known limitations (Tier 1 scope, by design)

• One document at a time, no cross-document search yet
• No auth, no UI, no streaming
• Fixed-size chunking with overlap, not structure-aware
• Plain vector similarity only - no reranking or hybrid search