CardSavvy

Never miss cashback again. CardSavvy chooses the best card for you.

Comment

Inspiration

Many people own multiple credit cards but still use the wrong one for purchases. Rewards are confusing, category-based, and hard to track. Most users realize they missed cashback only after the transaction is complete.

We wanted to build a system that brings decision intelligence before spending, not after redemption.

What It Does

CardSavvy recommends the best credit card for every purchase.

Users can:

  • Add their credit cards to their wallet
  • Enter a merchant name and transaction amount
  • Instantly see the optimal card
  • View estimated cashback and a clear reward explanation

Unknown cards are never auto-trusted. If reward information is extracted from the web, the card is stored as pending until user confirmation.

This ensures reliability and prevents incorrect reward assumptions.

How We Built It

We built the MVP using Mocha for rapid prototyping.

The system follows this architecture:

  • Use deterministic merchant-category mapping for analysis
  • Apply deterministic, rule-based reward logic
  • Calculate expected cashback across verified wallet cards
  • Recommend the highest-return option
  • Use Gemini for chatbot reasoning and unknown-card web extraction with confirmation + pending verification

Merchant detection and reward calculation are fully deterministic to ensure financial correctness.

AI is used only for:

  • Chat-based reasoning
  • Natural language interaction
  • Unknown-card lookup and structured extraction

This prevents hallucinated financial math and keeps calculations transparent.

Challenges We Ran Into

Balancing AI flexibility with financial accuracy was our biggest challenge.

We deliberately separated:

  • Deterministic reward computation
  • AI-powered reasoning and interaction

Another challenge was designing a verification layer so that web-extracted card data is never automatically trusted.

Accomplishments That We're Proud Of

  • A working MVP with real-time optimization
  • Transparent and explainable reward calculations
  • A trust-first architecture with pending verification
  • Clean chatbot interaction powered by Gemini

What We Learned

We learned that optimization at the moment of decision is far more powerful than post-spend tracking.

We also learned that AI works best when paired with deterministic systems in financial applications.

What's Next for CardSavvy

  • Integrate transaction aggregation providers (Plaid / Open Banking)
  • Automatically ingest real transaction data
  • Add regret tracking and reward analytics
  • Expand evaluation and observability for smarter optimization

Built With

Share this project:

Updates