← Back to Projects
🔗

Advanced RAG Chatbot

GitHub ↗
The Business Problem

Most RAG demos return an answer with no way to tell if it's actually grounded in the source documents — teams can't trust them enough to put in front of real users.

The Solution

A production-grade RAG chatbot that scores every response for faithfulness and relevancy before it reaches the user, with re-ranking to cut irrelevant answers.

The Result
  • Cross-encoder re-ranking reduces irrelevant-answer rate versus single-stage retrieval
  • Every response scored on faithfulness, relevancy, precision, and recall before delivery
  • Sliding-window session memory keeps multi-turn conversations coherent
  • RAGAS-inspired evaluation suite gives an ongoing quality signal, not a one-time check
Architecture

User Query -> Bi-Encoder Retrieval -> Cross-Encoder Re-Ranking -> LLM Generation -> Faithfulness Scoring -> Response

⚙️ Technical Details For engineers & technical reviewers — click to expand

Cuts irrelevant-answer rate by adding a cross-encoder re-ranking stage most RAG demos skip entirely — every response is scored on faithfulness and relevancy before it reaches the user, the kind of observability a production team actually needs. A complete Retrieval-Augmented Generation system built from scratch: raw document ingestion, sliding-window conversation memory with LLM context injection, and a RAGAS-inspired evaluation suite. Built to highlight real-world architectural decisions, not prototype-level RAG.

RAG Cross-encoder re-ranking Sentence-transformers ChromaDB Vector search Flask Session management RAGAS evaluation Faithfulness scoring OpenAI · Ollama

Need something like this for your own business?

Build Something Similar →

Have questions about this project? Ask my AI assistant for details.

Ask AI about this →