Veritas AI — Grounded RAG Document Intelligence System

A production-grade Retrieval-Augmented Generation (RAG) system engineered to eliminate LLM hallucinations by enforcing strict evidence grounding, page-level metadata tracking, and verbatim source citations.

🚀 Key Features

• Strict Source Grounding: Rejects queries outside document scope instead of guessing.
• Page-Level Citations: Pinpoints the exact file and page number for every claim made.
• Evidence Verification: Displays top retrieved chunks alongside cosine similarity match percentages.
• Local Embedding Pipeline: Computes high-performance dense representations on-device via sentence-transformers/all-MiniLM-L6-v2.
• Low-Latency Inference: Uses Groq Cloud LPU inference for high-speed generation.
• Lightweight Architecture: Vanilla JavaScript frontend with zero build toolchain overhead + FastAPI asynchronous backend.

🛠️ Architecture & Tech Stack

• Backend: FastAPI (Python 3.11)
• Vector Database: ChromaDB
• Embeddings: HuggingFace sentence-transformers/all-MiniLM-L6-v2 (384-dimensional dense vectors)
• LLM Engine: Groq API (gemma2-9b-it / llama-3.3-70b-versatile)
• Orchestration: LangChain
• Frontend: HTML5, CSS3, Vanilla JavaScript (XSS-safe DOM construction)

📦 Setup & Installation

1. Clone the repository

git clone [https://github.com/](https://github.com/)<your-username>/veritas-ai.git
cd veritas-ai