E-Commerce Support Agent

My first AI-powered customer support agent for e-commerce, built with Gemini and LangChain.

This agent uses Retrieval-Augmented Generation (RAG) to provide grounded answers from the Hugging Face E-Commerce Customer Support QA dataset. It demonstrates how to build a vector index, retrieve relevant knowledge, and generate concise answers using a modern LLM.

---

Installation

1. Clone the repository

   git clone https://github.com/eershr/e-commerce-support-agent.git
   cd e-commerce-support-agent
   

2. Install dependencies

   pip install -r requirements.txt

3. Build the index and run the agent

   python scripts/build_index.py
   python scripts/run_qa.py

Example:

Project Overview & Learnings

Dataset:

• Used the Hugging Face E-Commerce Customer Support QA dataset in parquet format.

Embeddings:

• Chose text-embedding-004 for semantic embeddings.
• Why text-embedding-004? Default choice for new RAG pipelines in 2025.
• Provides better semantic clustering, multilingual support, and domain generalization.

Vector store choice:

• Used FAISS instead of Chroma.
• FAISS is fast, memory-efficient, and ideal for large datasets.
• Ease of integration into local environments and AI frameworks like LangChain.

Pipeline choice:

• Used RetrievalQA instead of a fully Runnable pipeline.
• Simpler to implement for question-answering.
• Runnable pipelines provide more flexibility but require more setup and are better for multi-step reasoning or custom transformations.

Other key learnings:

• Prompt design is critical — adding instructions like “Answer only using the provided context” significantly improves answer quality.
• FAISS retrieval + structured Q&A documents ensures the agent rarely answers “I don’t know.”