FraudFrog

Inspiration

Traditional fraud monitoring is where efficiency goes to die. Risk analysts are routinely drowned in high-density, multi-column spreadsheets, forced to evaluate complex data structures under extreme time constraints. This fatigue leads to critical oversights: missing a malicious actor costs thousands in chargeback fees, while over-correcting blocks legitimate buyers and ruins customer trust.

What it does

FraudFrog re-engineers trust and safety workflows by replacing dense database grids with an ultra-fast, interactive "Tinder-style" triage queue.

After ingesting an enterprise transaction spreadsheet, our multi-layered engine processes the ledger and outputs a clean, deterministic threat score. Safe logs route to the front for rapid-fire approvals, while highly critical anomalies go to the back for aggressive damage control. By mapping operations to simple touch swipes or keyboard hotkeys (A to Approve, D to Reject, E to Escalate), analysts can clear hundreds of backlogs in a fraction of the time without facing cognitive fatigue.

How we built it

FraudFrog was engineered utilizing a premium, high-contrast fintech aesthetic powered by a full-stack Next.js architecture:The Stack: Built using Next.js 15 (App Router), TypeScript, and Tailwind CSS, utilizing interactive 3D stack components and Framer Motion for micro-interactions.The Rules Engine: Before reaching the UI, an automated heuristic pipeline scores incoming transactions cleanly between $0$ and $100$. The algorithm computes an accumulated risk coefficient ($R$) based on independent penalty weights ($w_i$) assigned to triggered heuristic rules ($x_i$): $$R = \min\left(\sum_{i=1}^{n} w_i x_i, \; 100\right)$$

Where $x_i \in {0, 1}$ represents rule violations like extreme standard deviation from a cardholder's median spend or suspicious cross-card device pooling. The final output dynamically segments items into three core UX actions:

$$\text{Operational Zone} = \begin{cases} \text{Green (Auto-Approve / Swipe Right)} & 0 \le R \le 15 \ \text{Amber (AI Triage / Swipe Up)} & 16 \le R \le 80 \ \text{Red (Loss Prevention / Swipe Left)} & 81 \le R \le 100 \end{cases}$$

Challenges we ran into

State & Animation Synchronization: Building a queue optimized for high-speed swiping caused major race conditions early on. Rapid hotkey inputs triggered state updates before Framer Motion could finish executing its exit paths, causing cards to clip or get stuck. We had to implement synchronized boundary locks within our custom React hooks to isolate array shifts and preserve fluid animations.

The 11-Column UI Clutter: Fitting an entire 11-column dataset onto a sleek interface was a massive UX challenge. Dumping every IP subnet and device string rendered the cards unreadable. We solved this by adopting a rigorous progressive disclosure model: high-signal data points (amounts and merchants) live on the front face, user baselines sit pinned to a contextual sidebar, and deep technical footprints stay elegantly hidden on the back of the card until the analyst clicks to flip it in 3D space.

Accomplishments that we're proud of

The Gemini AI Amber Triage Core: Instead of forcing human reviewers to manually debate ambiguous medium-risk (Amber Zone) logs, we successfully engineered an asynchronous Gemini AI Triage Agent. The agent intercepts mid-tier items, securely pulls the transaction metadata, evaluates it against a stringified array of that cardholder's past legitimate history, and appends a direct risk recommendation alongside a clear, 1-sentence reasoning to the back of the card.

Network-Wide Session Immunization: We built a live feedback loop. When an analyst swipes left to mark a critical item as definitive fraud, our system instantly extracts its unique device_id and ip_address, automatically shifting the threat threshold for any remaining cards in the deck that share those identical parameters.

What we learned

Building FraudFrog emphasized the massive power of human-in-the-loop AI systems. In high-stakes fintech environments, AI shouldn't completely replace manual human intuition. Instead, its true purpose is to eliminate noise, manage information architecture, and empower human analysts to make incredibly accurate, cost-aware decisions at maximum velocity.

What's next for FraudFrog

We plan to scale our cross-card network aggregations into a decentralized ledger, allowing independent merchants using FraudFrog to instantly share blacklisted digital fingerprints globally. We also want to integrate real-time API webhooks, transforming our static CSV-triage tool into a fully live, inline payment gate guard dog.