Cart guardian

Buy with clarity. CartGuardian uses AI to help you make smarter purchasing decisions, understand the risks and know the true cost before you checkout.

Comment

CartGuardian - Buy with clarity.

Inspiration

We've all experienced that moment before making a purchase: finding something we really want, looking at the price, and wondering whether it's actually worth it.

Is the product worth the money? Are there hidden costs? Does it meet our needs? Are we making an informed decision, or simply giving in to the excitement of buying something new?

We wanted to build something that helps people pause and think before they spend.

That's how CartGuardian was born: an AI-powered purchase evaluation platform designed to give consumers a second opinion before checkout.

Our idea was simple: instead of letting AI make purchasing decisions for people, why not use it to help people make better-informed decisions themselves?

What it does

CartGuardian helps users evaluate products before committing to a purchase.

Users can either paste a product URL to automatically import available product information or manually enter details for products they're considering buying in person.

Using Gemini AI, CartGuardian evaluates the purchase and presents an understandable recommendation:

  • Worth considering
  • Proceed with caution
  • Not recommended

But we wanted to go beyond a simple AI verdict.

Our Purchase Confidence Card gives users:

  • A qualitative AI confidence rating.
  • The strongest arguments for and against the purchase.
  • Evidence labels distinguishing user-provided information, general knowledge and assumptions.
  • Financial calculations such as cost per use and cost per month.
  • Visible uncertainties and missing information.
  • Practical questions and checks to consider before purchasing.

Users can inspect the full evaluation, edit imported product details and remain in control of their final decision.

When checkout is supported, they can proceed through PayPal Sandbox.

How we built it

CartGuardian was built using a full-stack TypeScript architecture.

Our technology stack includes:

  • React 19 + Vite: Frontend application.
  • TypeScript + Express 5: Backend API.
  • Gemini AI: Purchase evaluation and reasoning.
  • PayPal Orders API + JavaScript SDK: Checkout integration.
  • Supabase + PostgreSQL: Persistent data storage.
  • Zod: Shared validation and data contracts.
  • Vercel: Production deployment.

One of our most important architectural decisions was separating AI-generated assessments from deterministic application logic. The AI provides the purchase evaluation, while financial calculations and payment operations are handled separately.

We also developed a product-link import system that extracts publicly available product information using structured metadata and product data embedded in retailer pages. Imported information remains editable, and manual entry is always available.

Because accepting arbitrary URLs introduces security risks, we implemented URL validation, internal network protections, response-size limits, request timeouts and rate limiting.

Challenges we ran into

Making AI recommendations trustworthy

We didn't want CartGuardian to confidently present assumptions as facts. We introduced evidence provenance labels, qualitative confidence ratings and visible uncertainty so users can understand where information comes from.

Extracting information from retailer websites

Retailers expose product information differently, and some restrict automated access. We built a layered extraction system and tested it against multiple real retailers.

Takealot presented a particular challenge because its product pages are protected by Cloudflare. Rather than attempting to bypass those protections, we explain the limitation and provide manual entry as a fallback.

Payment and currency limitations

Integrating PayPal involved implementing authentication, order creation, approval and capture. We also encountered a regional limitation: PayPal checkout does not support ZAR in our implementation. CartGuardian therefore supports evaluating rand-priced products while clearly explaining why checkout is unavailable.

Production deployment

Getting our full-stack application running reliably on Vercel required us to address build compatibility and serverless execution challenges. We introduced Supabase persistence and atomic capture-claim logic to ensure that important order operations weren't dependent on temporary in-memory state.

Accomplishments that we're proud of

We're particularly proud that CartGuardian progressed beyond a local prototype into a working, publicly accessible application.

Some of our biggest accomplishments include:

  • Successfully deploying CartGuardian to Vercel.
  • Integrating Gemini AI for purchase evaluations.
  • Completing a real end-to-end PayPal Sandbox transaction on the live website.
  • Verifying that the completed transaction was persisted in Supabase with a COMPLETED order status.
  • Building product-link imports tested against real retailers, including Ridge, Fellow, Espresso Parts, Checkers, Evetech, Computer Mania, Faithful to Nature and Loot.
  • Supporting both automated product imports and manual product entry.
  • Building a transparent Purchase Confidence Card that distinguishes AI assessments from financial calculations.
  • Implementing security protections around external product URL retrieval.
  • Achieving 187 passing automated tests.

Most importantly, we're proud that we built something that doesn't simply tell users what to buy. It helps them understand the reasoning behind a purchasing decision.

What we learned

This project taught us that building a meaningful AI application involves much more than connecting an API to a frontend.

We learned how important it is to distinguish AI-generated reasoning from factual information, design for uncertainty, validate external data and build reliable payment workflows.

We also gained practical experience with serverless deployment, persistent database architecture, payment integration and secure web data extraction.

Perhaps our biggest lesson was that a trustworthy AI system should be comfortable acknowledging what it doesn't know.

We believe AI should empower people with information and reasoning, not take away their ability to make their own decisions.

What's next for CartGuardian?

CartGuardian is a foundation for more transparent and informed purchasing decisions.

Looking ahead, we'd like to:

  • Expand support for additional retailers and product categories.
  • Introduce side-by-side product comparisons.
  • Improve long-term cost analysis and purchase planning.
  • Expand currency and payment support for more regions.
  • Continue improving evaluation accuracy and transparency.

Our vision is to make CartGuardian a trusted companion that people can consult before making important purchases.

We don't want to tell people what they should buy.

We want to help them understand what they're buying, why it might or might not be worth it, and what they should consider before spending their money.

CartGuardian - Buy with clarity.

Built With

Share this project:

Updates

Submission history