Work / AI prototype · Team of 3 · 2025
FinSightcards, sorted
A conversational credit-card recommender, built in a two-week sprint. It learns who you are in 15 questions, then matches you to a card and explains why.
Watch it work
The real app, 75 seconds: sign in, answer the 15 questions, get the match.
The problem
Choosing a credit card is confusing. Cards differ on many small details, and most comparison sites rank by who pays them, not by what suits you.
What FinSight is
A chat that learns who you are, then recommends the one card that fits you and explains why. It is built for people who do not read fine print.
- Sign in with email or Google.
- Dashboard: "Smarter cards. Sharper choices." You choose Find My Match.
- 15 questions: age, job, income, city, current cards, credit score, spending, bank preferences.
- The match: the best-fit card with reasons, key benefits, bonus, eligibility and a link to apply.
The data underneath
A structured database of 300+ credit cards, each with 8+ attributes. That is 2,400+ data points, organised so the app can compare a person against every card.
How it scores
A multi-parameter framework that scores each card on 8 financial and usage criteria, then weights them by what the person told us. The model explains the result in plain English, so the answer is never a black box.
How we built it
- Two-week sprint with a team of three.
- Lovable for the app, Supabase for the database and search, 4+ development cycles.
- Prompt design and token efficiency: shorter, sharper prompts, because every extra word costs money at scale.
Testing and the judges
Tested with 25+ users, then presented the working prototype to 2 judges.
What I'd do next
- Pull live card data through an API instead of a manual database.
- Learn the data and API layer properly. That is why I am learning Python now.