RAG-Powered Document Assistant

Axiom RAG is an enterprise-grade Retrieval-Augmented Generation system designed for grounded question answering across academic computer science literature and research papers. Operating 100% locally with zero cloud dependencies, the architecture eliminates hallucination through dense semantic embeddings, persistent vector indexing, and zero-shot grounding guardrails that enforce verbatim page-level citations.

The Core Problem: Standard Large Language Models frequently invent convincing but factually incorrect citations, misquote algorithms, and require expensive API tokens. Furthermore, academic research institutions and enterprise clients cannot upload proprietary research or confidential internal documentation to public cloud models.

The Engineering Solution: Constructed a robust offline RAG pipeline using Sentence Transformers (all-MiniLM-L6-v2) to generate dense vector embeddings, ChromaDB for high-throughput persistent vector indexing, and Ollama to orchestrate local quantized models (LLaMA 3 / Mistral). The prompt architecture strictly penalizes ungrounded speculation and forces the LLM to output source page numbers and exact passage citations.