Back to portfolio GitHub

Amazon Rufus-inspired product assistant

AI Shopping Agent

A conversational shopping assistant that uses LLM tool calling to search products, inspect ratings and reviews, recommend options, and support shopping actions through a Streamlit interface.

End-to-end tool-calling workflow with a persistent product data layer
LangChainOpenAIStreamlitSQLitePython

The challenge

Why this problem needed a different architecture

Shopping questions require the model to retrieve structured product data, apply constraints, explain tradeoffs, and take actions instead of simply generating text.

The solution

How I approached it

I connected an LLM to product, review, cart, and order tools, then wrapped the workflow in an interactive application with persistent SQLite storage.

Architecture

  1. Natural-language request interpreted by the agent
  2. Tool calls search structured product and review data
  3. Results are returned to the model as observations
  4. The model synthesizes a recommendation and next action
  5. Streamlit provides a conversational front end

Results

  • Supports multi-constraint product discovery
  • Demonstrates explicit Think → Act → Observe agent behavior
  • Connects conversational AI with structured transactional data