Flex

Search once. Compare everywhere.

Flex is a web app that lets you search for a product and compare its price, rating, and reviews across Amazon India, Flipkart, and Myntra, with a short AI-generated summary of what buyers like and don't like.

What's real vs. demo data (important)

This is the Phase 1 MVP, running in demo mode:

• Merchant data (prices, ratings, reviews) is seeded sample data, not live Amazon/Flipkart/Myntra listings. None of those platforms offers a free, individually-accessible API, and scraping them would violate their own terms of service and robots.txt — see docs/ARCHITECTURE.md for the details and docs/ROADMAP.md for the legitimate path to real data later.
• "View Deal" links are real. They go to each platform's own public search-results page for the matched product — not scraped, just a constructed URL — so clicking through actually works today.
• The AI Review summary is a deterministic, rule-based summarizer, not a real LLM call. It only ever uses the review text it's given — it can't invent anything beyond that.
• The database is a real Supabase Postgres project.

Running it locally

You need Node.js 20.9 or later installed.

1. Install dependencies:

   npm install
   

2. Copy .env.example to .env and fill in a real DATABASE_URL (see "Database setup" below). This is the only credential Flex needs — no merchant or AI API keys required for Phase 1.
3. Start the app:

   npm run dev
   

4. Open http://localhost:3000 and try searching "iPhone 17 Pro 256GB" or any of the examples on the homepage.

Database setup

Flex uses Supabase's free Postgres tier.

1. Create a free Supabase project (or use the one already provisioned for this project).
2. In the project dashboard, go to Project Settings → Database → Connection string and copy the direct connection URI (port 5432, not the pooled one).
3. Paste it into .env as DATABASE_URL.

If the database is ever unreachable, Flex still works — price/rating caching degrades gracefully to "compute fresh" instead of crashing the page.

Testing

npm test          # unit tests (Vitest) — normalization, matching, providers, validation, AI safeguards
npm run test:e2e  # end-to-end smoke test (Playwright) — home → search → compare → AI summary → redirect

Project layout

See docs/ARCHITECTURE.md for the full picture. The short version:

app/                   Pages and routes (Next.js App Router)
lib/core/providers/     Amazon/Flipkart/Myntra adapters (mocked in Phase 1)
lib/core/matching/      Turns messy merchant titles into comparable products
lib/core/ai/            AI review summarizer (mocked in Phase 1)
lib/core/search/        Ties providers + matching + caching + AI together
lib/db/                 Prisma client and caching logic
lib/security/           Rate limiting and the merchant-redirect allowlist
components/             UI components

More docs

• docs/ARCHITECTURE.md — how the pieces fit together, with a diagram
• docs/SECURITY.md — threat model, mitigations, and what's still open
• docs/ROADMAP.md — what's built, what's next, what's deliberately deferred