PRJ_08 · backend
Document Intelligence Service
A RAG pipeline that turns a document archive into an AI-queryable knowledge base.
The problem
A large archive of scanned and native PDFs had no way to search or query by content, only by manually opening files one at a time.
The solution
Two independently-deployed FastAPI services: an async job-based ingestion pipeline (PyMuPDF extraction with Tesseract OCR fallback, text normalization, chunking, multilingual sentence-transformer embeddings into pgvector) feeding a retrieval-augmented question-answering and summarization layer, with an output-validation stage that checks every LLM response for schema conformity, citation accuracy, PII leakage, and hallucination before it reaches a user.
Have a similar problem?
I'm open to remote roles and freelance work. Tell me what you're building.