The HTS AI Agent is a dual-function intelligent assistant designed to:

1. 🧠 Answer trade policy and agreement-related queries using RAG (Retrieval-Augmented Generation) on HTS General Notes documentation.
2. 💸 Compute duties and landed cost based on HTS code and CIF (Cost + Insurance + Freight) details using DuckDB and HTS tariff CSVs.

The agent leverages:

• ChromaDB + GoogleGenerativeAIEmbeddings for RAG
• DuckDB for HTS data storage and duty computation
• LangChain MultiTool Agent for smart routing between tools
• Streamlit for the frontend UI

graph TD
User["🧑 User (NL Query via Streamlit)"]
Preprocessor["🔍 Query Router"]
RAGAgent["📚 RAG Agent (Chroma + Gemini)"]
SQLAgent["🧾 SQL Agent (DuckDB + Duty Calculator)"]
ResponseFormatter["🖼 Markdown Output"]
CSVExporter["⬇️ Downloadable CSV (landed_cost_duties.csv)"]

User --> Preprocessor
Preprocessor --> RAGAgent
Preprocessor --> SQLAgent
RAGAgent --> ResponseFormatter
SQLAgent --> ResponseFormatter
ResponseFormatter --> CSVExporter

• Loads and chunks General Notes Full Documentation (PDF)
• Embeds using GoogleGenerativeAIEmbeddings
• Stores vectors in ChromaDB
• Performs semantic search for trade-related questions

• Loads Section I CSV into DuckDB
• Accepts HTS code + product cost, weight, quantity
• Parses duty strings (% / ¢ per kg / $ per unit)
• Computes all applicable duty amounts and total landed cost
• Returns breakdown in Markdown and CSV format

• Dynamically chooses between RAG and SQL tools
• Supports multi-tool queries in one prompt
• Markdown-rich answers with data tables
• Exportable CSV of calculated duties

• What is the United States-Israel Free Trade Agreement?
• Can a product that exceeds its tariff-rate quota still qualify for duty-free entry under GSP or any FTA?
• How is classification determined for an imported item that will be used as a part in manufacturing but isn’t itself a finished part?

• HTS Code: 0101.30.00.00, Cost: $10,000, Weight: 500kg, Units: 5
• What’s the HTS code for donkeys?
• What are the applicable duty rates for female cattle?

1. Install Dependencies (via uv):

   uv pip install -r requirements.txt

2. Run the Streamlit App:

    streamlit run .\src\frontend\app.py

3. Interact:

   • Enter a trade policy question or HTS-based duty calculation request.
   • View the Markdown output and download CSV (if applicable).

• ✅ GitHub Repository (Codebase)
• ✅ Demo Video (RAG + SQL Agent in action)
• ✅ This README
• ✅ Resume of Ritwik Singh

Thanks to Personaliz.ai for this unique opportunity to blend AI, data extraction, and international trade policy into one intelligent assistant.