RAG Lens is a high-fidelity observability tool designed to strip away the "black box" of Retrieval-Augmented Generation. While most RAG applications only show the final output, RAG Lens allows engineers to visualize the entire data lifecycle—from semantic chunking to verified generation.

• Dual-Strategy Chunking: Compare standard Fixed-Size character splitting against Semantic Chunking (AI-driven sentence similarity) to see how data structure affects retrieval.
• Retrieval Transparency: Side-by-side UI that displays the exact document chunks retrieved from the vector store before they are passed to the LLM.
• Hallucination Guard: Implements a multi-agent "Fact-Checker" pattern that uses a second LLM pass to verify the final response against the source context.
• Enterprise-Ready Security: Secure credential management with an automated fallback system (Streamlit Secrets vs. User-provided keys).
• Instant Onboarding: Includes a pre-loaded technical dataset on RAG Best Practices for immediate testing and demonstration.

• Orchestration: LangChain
• LLM Inference: Groq (Llama 3 70B)
• Embeddings: HuggingFace all-MiniLM-L6-v2
• Vector Database: FAISS (Local CPU)
• Frontend: Streamlit

1. Clone the repository

   git clone [https://github.com/your-username/rag-lens.git](https://github.com/your-username/rag-lens.git)
   cd rag-lens

2. Install dependencies

   pip install -r requirements.txt

3. Configure Secrets Create a .streamlit/secrets.toml file and inside the file add your Groq API key:

   GROQ_API_KEY = "your_grok_api_key_here"

4. Run the app

   streamlit run app.py

As a senior engineer, I realized that the biggest challenge in production AI isn't just getting an answer—it's getting a reliable answer. I built RAG Lens to provide a debugging suite that helps developers:

1. Visualize how different embedding models "see" their data.
2. Debug "Noise in Context" by seeing exactly what chunks are retrieved.
3. Minimize hallucinations using an automated verification layer.

This project uses .gitignore to ensure that local .streamlit/secrets.toml files are never committed to version control, following DevSecOps best practices.