IntelliQuery - Local RAG (Ollama)

Local RAG system over PDFs with conversation memory, hybrid search, re-ranking and citation-grounded answers.

• Pipeline: PDF ingestion → chunking → embedding (Ollama) → Chroma vector store → retrieval → rerank → LLM generation → strict fallback

Features

• PDF ingestion with chunking
• Dense + hybrid retrieval (BM25 + embeddings)
• Multi-turn memory
• Query normalization & expansion
• Sentence-level citations
• Retrieval confidence & grounding validation
• Streamlit chat UI with persistent history

Metrics: Precision@k, Recall@k, MRR, Answer Relevance & Groundedness

Quick Start

pip install -r requirements.txt
ollama pull hf.co/CompendiumLabs/bge-base-en-v1.5-gguf
ollama pull hf.co/bartowski/Llama-3.2-1B-Instruct-GGUF
streamlit run app.py

• Upload PDFs → Build Index → Ask Questions

Next branch (planned)

• multi-modal-rag → add image/PDF-table ingestion