Skip to content

README.md

Latest commit

 

History

History
52 lines (39 loc) · 2.86 KB

README.md

File metadata and controls

52 lines (39 loc) · 2.86 KB
Raw

ShopBuddy

Problem statement

Online shopping decisions are opaque. ShopBuddy solves this by providing transparent reasoning.

Why multi-agent?

This workflow has multiple specialized tasks that are easier, safer, and more reliable when split across cooperating agents: query understanding, marketplace collection, quality filtering, ranking, review intelligence, and explanation generation. A single agent can answer quickly, but a multi-agent pipeline gives clearer accountability, better explainability, and cleaner separation of concerns.

Agent architecture

Agent Role
Input Safety Gate Agent Blocks harmful, non-shopping, or prompt-injection style user input before orchestration starts.
Query Interpreter Agent (Supervisor) Converts user intent into clean search terms, budget, and constraints; starts a search session.
Human Approval Checkpoint Pauses for user confirmation of extracted keywords before scraping begins.
Marketplace Collector Agent (Scraper) Fetches products from Daraz and Amazon, maps to a shared schema, and normalizes currency.
Quality Gate Agent (Filter) Applies relevance, budget, reviews, and duplicate filters with explicit pass/fail reasons.
Value Ranker Agent (Analyzer) Computes value scores and assigns recommendation badges to top candidates.
Review Analyst Agent (Reviewer) Summarizes customer review sentiment, themes, and trust signals for top products.
Explainability Narrator Agent (Explainer) Generates plain-English transparency reports and per-product recommendation reasoning.

State Graph Visualization:

Workflow Diagram

How to run

  1. Clone the repository.
  2. Backend setup:
    • cd backend
    • Create .env from backend/.env.example
    • pip install -r requirements.txt
    • uvicorn main:app --reload
  3. Frontend setup:
  4. Open the frontend URL shown by Vite (usually http://localhost:5173).

Tech stack

  • Languages: Python, JavaScript (React)
  • Backend: FastAPI, LangGraph, LangChain, SQLAlchemy
  • Frontend: React, Vite, Tailwind CSS, shadcn/magicui components
  • Models: Gemini 3 Flash, OpenAI gpt oss 20B
  • Data Sources: Daraz scraper, Amazon scraper
  • APIs: Google GenAI API, Groq API, ExchangeRate-API
  • Database: Supabase Postgres