FynX | Devpost

Inspiration

Pakistan has 65 million digital banking users and 5.2 million SMEs, but almost none of them have real financial visibility. People screenshot bank SMS alerts and forget them. Small business owners track sales in notebooks. And when SMEs apply for loans, 70–80% get rejected not because they're bad businesses, but because there's no data trail proving they're good ones. We built FynX because we kept watching people we knew including our own families make financial decisions blind, and we knew AI could fix that gap cheaper and faster than a human accountant ever could.

What it does

FynX is an AI-powered financial operating system with two apps under one roof.

FynX Personal automatically captures every transaction straight from bank SMS alerts — no manual entry and reads paper receipts through OCR. Gemini then turns that raw data into a Financial Health Score (0–100), a spending forecast, a subscription radar that catches recurring charges you forgot about, and savings vaults that nudge you toward goals.

FynX Business does the same for SMEs: automated bookkeeping, inventory, and payroll, plus CreditIQ an alternative credit-scoring engine that gives lenders a real picture of a business's health even without a formal credit history, and a Business CFO Chat that answers financial questions in plain language.

How we built it

FynX is built in Flutter and Dart with Clean Architecture and BLoC for state management, backed by Firebase and Firestore. Transaction capture runs through an Android BroadcastReceiver that listens for bank SMS in real time, and receipt scanning uses Google ML Kit for OCR. All the intelligence — health scoring, forecasting, CreditIQ, the CFO chat runs on a multi-model Gemini fallback chain, so the app stays responsive even if one model is under load. Local caching runs through Hive to keep the app fast offline.

Challenges we ran into

Parsing real-world bank SMS formats turned out to be far messier than expected — every bank writes its alerts differently, and we had to build a robust extraction and cleanup layer just to get consistent structured data out of free-text messages. We also hit real engineering friction getting Gemini's OCR pipeline to reliably read printed receipts versus handwritten ones, which meant iterating on model selection and prompt design until accuracy held up. On top of the product work, we were simultaneously prepping incubation and regulatory applications (SBP Regulatory Sandbox, NIC KHI, PSF), which meant building and fundraising in parallel on a tight timeline.

Accomplishments that we're proud of

We shipped a working dual-app architecture - personal and business finance powered end-to-end by AI rather than static rules. We built CreditIQ, which we believe is a genuinely novel approach to alternative credit scoring for underserved SMEs in Pakistan. And we did all of this while also completing a full pre-launch security and compliance audit, positioning FynX for a real Play Store launch rather than just a hackathon demo.

What we learned

We learned that AI-native financial tooling isn't just about bolting a chatbot onto a finance app it's about letting AI drive the core decisions: what gets flagged, what gets forecast, what gets scored. We also learned a lot about the regulatory landscape for fintech in Pakistan, including where FynX sits (analytics, not payments) and why that keeps us outside SBP licensing requirements for now.