TrustLedger

Financial fraud is becoming increasingly sophisticated, but understanding whether a transaction is suspicious remains difficult for everyday users. Traditional fraud detection systems often provide little explanation beyond a warning or a risk score, leaving users uncertain about what happened and what to do next. We wanted to build a financial safety assistant that doesn't just flag suspicious transactions—it helps people understand risks, make informed decisions, and learn from previous experiences without sacrificing control over their personal data. That idea became TrustLedger: an explainable, privacy-first AI financial safety assistant. What It Does TrustLedger combines transaction monitoring, fraud risk analysis, AI-generated explanations, and consent-based financial memory into one experience. Our system is designed around five core capabilities:

1. Transaction Monitoring: Integrates with Capital One's Nessie API to retrieve sandbox banking transactions and display them in a user-friendly dashboard.
2. Fraud Risk Detection: Analyzes transactions using deterministic risk rules and an extensible AI fraud-provider interface.
3. Explainable AI: Integrates Google Gemini to translate technical fraud signals into understandable explanations, helping users understand why a transaction might be suspicious.
4. Human-in-the-Loop Review: Keeps users responsible for confirming transaction legitimacy instead of allowing AI predictions to make irreversible decisions.
5. Privacy-First Financial Memory: Uses consent-controlled storage, including Backboard-powered memory integration, to retain relevant context from reviewed transactions while maintaining user ownership and separation. Our central philosophy is simple: AI should help users make financial decisions, not make those decisions for them.

Built With

• backboard
• fastapi
• geminiapi
• jev
• nessieapi
• python
• react
• sqlite
• typescript