🎓 GAC Salem - Advanced AI Campus Bot (RAG Enterprise Edition)

An intelligent, context-aware RAG (Retrieval-Augmented Generation) system built to dynamically handle complex institutional queries for Government Arts College Salem.

---

🌿 Branch Strategy & Project Evolution

This repository is intentionally structured to showcase the evolution of an AI product from prototype to production:

• main branch:
  Contains the foundational Generative AI chatbot built with basic scraping and LLM generation.

• advanced-rag-pipeline branch (Current):
  The enterprise-grade upgrade. This branch introduces strict multi-stage routing, vector-based semantic caging, and mathematical hallucination prevention to make the bot production-ready.

---

🚀 Key Enterprise Features

• Multi-Stage Micro-Bucket Routing:
  Uses an LLM to dynamically route user queries into 5 strict semantic categories (e.g., Department, PO-PSO-CO, General) before retrieval.

• Deterministic Inverse Filtering:
  Intelligently separates structural PDF documents (Syllabi/Course Outcomes) from Faculty Profiles without relying on standardized naming conventions.

• Optimized Token Payload:
  Passes mathematically filtered, highly relevant chunk pools (approx. 7,500 tokens) to guarantee zero "Lost in the Middle" attention degradation.

• Strict Lexical Immunity:
  Bypasses traditional Cross-Encoder reranker collapse by forcing structural list retention for department/course queries.

• LLM-as-a-Judge Evaluation:
  Includes an automated production evaluation pipeline using the RAGAS framework, mathematically proving:

  • 100% Faithfulness
  • 100% Answer Relevancy

  across synthetic user testing.

---

🛠️ Technology Stack

Component	Technology
Frontend	Streamlit
Orchestration	LangChain Core
Embeddings	Mistral AI (mistral-embed)
Vector Database	ChromaDB
Generator LLM	Azure OpenAI (gpt-4o-mini)
Evaluation Engine	RAGAS & Pandas

---

⚙️ Quick Start (Windows)

1. Clone and Checkout the RAG Branch

git clone https://github.com/Baskar-forever/college_chatbot.git
cd college_chatbot
git checkout advanced-rag-pipeline

---

2. Create and Activate a Virtual Environment

python -m venv venv
venv\Scripts\Activate.ps1

---

3. Install Dependencies

pip install -r requirements.txt

---

4. Configure Environment Variables

Create a .env file in the project root:

AZURE_OPENAI_API_KEY=your_azure_key
AZURE_OPENAI_ENDPOINT=your_azure_endpoint
AZURE_OPENAI_API_VERSION=2024-12-01-preview
MISTRAL_API_KEY=your_mistral_key

---

5. Run the Application

streamlit run app.py

---

📊 Running the Production Evaluation

To validate the chatbot against hallucinations and context drop-off, run the automated RAGAS evaluation pipeline:

python production_eval.py

This will:

• Process a synthetic batch of evaluation queries
• Grade responses using mathematical RAG metrics
• Export a detailed audit log as:

production_evaluation_results.csv

---

🔀 Managing Multiple Versions Locally

Since both the simple chatbot and enterprise RAG system exist in separate branches, switch between them using:

Work on the Simple Version

git checkout main

Work on the Enterprise RAG Version

git checkout advanced-rag-pipeline

Your local files will automatically update to match the selected branch.

---

✅ Project Goal

This architecture is designed to solve the three major production RAG problems:

1. Hallucination
2. Context Collapse
3. Retriever Noise

The system achieves this through deterministic retrieval control, semantic routing, and evaluation-driven validation.