Agentic RAG for E-commerce

An Agentic Retrieval-Augmented Generation (RAG) system for e-commerce product search, combining intelligent agents, vector retrieval, and LLM-based evaluation (LLM as Judge).

Overview

This project implements a RAG pipeline that:

• Stores and retrieves product data using ChromaDB and OpenAI embeddings
• Uses a LangChain Agent to reason, reformulate queries, and generate natural answers to user queries
• Evaluates its own performance via an LLM Judge powered by GPT-4.1-nano (LLM as Judge)

---

Tech Stack

• Python – Core language
• LangChain – Agent framework
• ChromaDB – Vector store
• OpenAI API – Embeddings + LLMs (Agent + Judge)
• Rich / Pandas – UI + data handling

---

🚀 Quick Start

# 1. Install dependencies
pip install -r requirements.txt

# 2. Add your API key
echo "OPENAI_API_KEY=sk-xxx" > .env

# 3. Ingest product data
python src/ingest.py

# 4. Run the agent
python src/agent.py --query "Find Samsung smartphones under $1000"

# 5. Evaluate results
python src/evaluate.py --max-tests 5

---

Example Evaluation Output

Agentic Evaluation Summary
─────────────────────────────────────────────
Overall Score       0.82   ✅ Excellent
Relevance           0.72   ✅ Good
Completeness        0.90   ✅ Excellent
Structure           0.95   ✅ Excellent
Tone                1.00   ✅ Perfect
Accuracy            0.90   ✅ Excellent
Response Time       9.8s   ⚠️ Needs Optimization
─────────────────────────────────────────────
Status: 🚀 Production-Ready

---

Project Structure

ecommerce-agentic-rag/
├── data/
│   ├── products.csv
│   └── test_questions.csv
├── src/
│   ├── ingest.py
│   ├── agent.py
│   └── evaluate.py
└── requirements.txt