Local research paper QA system (RAG)

ConstellAI

Accurate, hallucination-resistant QA

Developed an end-to-end retrieval-augmented generation pipeline for querying local research papers with high factual accuracy.

PythonChromaDBsentence-transformersOllamaUMAP

Build Highlights

Built PDF ingestion pipeline with PyMuPDF and custom font-shift decoding.

Generated semantic embeddings using BAAI/bge-small-en-v1.5 and implemented top-k retrieval.

Ran Mistral locally via Ollama with strict source-grounded prompting.

Visualized embedding space with UMAP to create a 2D knowledge graph of documents.