The AI-native shopping assistant that actually understands your customers.
Powered by Google Gemini · LangChain Agents · Semantic Routing · RAG
TL;DR — ShoppingGPT is a production-ready, agentic AI chatbot for e-commerce. It routes every conversation through a semantic classifier, dispatches to specialized LangChain tools, and streams back intelligent responses — all in under a second.
Generic chatbots give generic answers.
Your customers ask "Do you have a red linen shirt in size M under $40?" and get "Please visit our website."
ShoppingGPT changes that. It understands intent, queries your live inventory with natural language, and replies like a seasoned sales associate — every single time.
| Capability | Detail |
|---|---|
| 🧠 Gemini 1.5 Flash LLM | Google's fastest production model for blazing response times |
| 🔀 Semantic Router | Classifies queries into product or chitchat in milliseconds |
| 🔍 NL → SQL Product Search | Converts plain English into optimised SQLite queries |
| 📜 Policy RAG (FAISS) | Retrieves relevant policy docs via vector similarity search |
| 💬 Conversation Memory | Full turn-by-turn context so nothing gets forgotten |
| 🌐 Multi-language | Responds in the same language the customer uses |
| ⚡ Flask REST API | Simple /get endpoint — drop into any frontend |
| 🔌 LLM-Agnostic | Swap Gemini for Groq / Llama3 with one line change |
Customer Message
│
▼
┌─────────────────────┐
│ Semantic Router │ ← HuggingFace embedding + cosine similarity
└──────┬──────────────┘
│
┌────┴────┐
│ │
▼ ▼
Chitchat Shopping Agent (LangChain ReAct)
Chain │
│ ┌───┴───────────────┐
│ │ │
│ ▼ ▼
│ Product Search Policy Search
│ (NL → SQL → (FAISS Vector
│ SQLite DB) Similarity)
│
└──────────────────────────────────▶ Response
+ Memory Update
- Semantic Router over keyword matching — intent classification generalises to unseen phrasing.
- NL → SQL via LLM — no rigid query parser; handles complex multi-filter queries out of the box.
- FAISS for policies — sub-millisecond similarity search; no database round-trip.
- Shared
ConversationBufferMemory— both the chitchat chain and shopping agent share context, so users don't have to repeat themselves.
git clone https://github.com/yourusername/ShoppingGPT.git
cd ShoppingGPTpython -m venv venv
# macOS / Linux
source venv/bin/activate
# Windows
venv\Scripts\activatepip install -r requirements.txtcp .env.example .env
# Open .env and paste your Google AI Studio API keyGOOGLE_API_KEY=your_google_api_key_hereGet your free key at aistudio.google.com
Web App (Flask)
python app.py
# → http://localhost:5000CLI Mode
python main.pyShoppingGPT/
├── app.py # Flask web server + request handler
├── main.py # Interactive CLI entry-point
├── requirements.txt
├── .env.example # Copy → .env and fill in keys
│
├── shoppinggpt/
│ ├── agent.py # LangChain ReAct Shopping Agent
│ ├── chain.py # Chitchat LLM chain
│ ├── config.py # Centralised config & path resolution
│ ├── router/
│ │ └── lib_semantic_router.py # Semantic routing logic
│ └── tool/
│ ├── product_search.py # NL → SQL → SQLite tool
│ └── policy_search.py # FAISS policy retrieval tool
│
├── data/
│ ├── products.db # SQLite product catalogue
│ ├── policy.txt # Store policy document
│ └── datastore/ # FAISS index (auto-generated)
│
└── templates/
└── index.html # Chat UI
Products are stored in SQLite with the following schema:
| Column | Type | Description |
|---|---|---|
product_code |
TEXT | Unique SKU identifier |
product_name |
TEXT | Full product name |
material |
TEXT | Fabric / material composition |
size |
TEXT | Available sizes (S, M, L, XL…) |
color |
TEXT | Available colours |
brand |
TEXT | Brand or manufacturer |
gender |
TEXT | Target gender (male / female / unisex) |
stock_quantity |
INTEGER | Units in stock |
price |
REAL | Price in USD |
Open main.py (or app.py) and uncomment:
# Gemini (default)
LLM = ChatGoogleGenerativeAI(temperature=0, model="gemini-1.5-flash")
# Groq / Llama3 (faster, free tier available)
from langchain_groq import ChatGroq
LLM = ChatGroq(temperature=0, model="llama3-8b-8192")Replace data/products.db with your own SQLite database (keep the schema above).
Edit data/policy.txt — the FAISS index rebuilds automatically on next run.
Modify ShoppingAgent.SYSTEM_PROMPT in shoppinggpt/agent.py.
You: Do you have any blue denim jackets?
ShoppingGPT: Yes! We have 3 blue denim jackets in stock:
• Levi's Classic Denim Jacket — M, L, XL — $59.99
• H&M Slim Denim Jacket — S, M — $34.99
• Wrangler Heritage Jacket — L, XL, XXL — $74.99
Would you like more details on any of these?
You: What's your return policy?
ShoppingGPT: We offer a hassle-free 30-day return policy on all items.
Products must be unworn with original tags attached...
You: It's raining today 😅
ShoppingGPT: Perfect weather to stay stylish! Have you seen our new
waterproof trench coat collection? They're a bestseller this season ☔
Clean, minimal chat UI
Real-time product search results
Context-aware casual conversation
Pull requests are welcome! Here's how to contribute:
# 1. Fork the repository
# 2. Create your feature branch
git checkout -b feature/your-feature-name
# 3. Commit with a meaningful message
git commit -m "feat: add cart management tool"
# 4. Push and open a PR
git push origin feature/your-feature-namePlease follow Conventional Commits and update tests where applicable.
The custom text classification model powering the semantic router:
hang1704/opendaisy on Hugging Face
Fine-tuned to classify queries as chitchat or product — feel free to fork it.
MIT © 2024 — see LICENSE for details.
- LangChain — the agent orchestration backbone
- Google Generative AI — Gemini LLM + embeddings
- Semantic Router — blazing-fast query routing
- FAISS — vector similarity at scale
Built with ❤️ · ShoppingGPT — Because your customers deserve better than keyword search.