FraudShield AI — Evidence-First Agent

Ins## Inspiration

Fraud investigations require explainable decisions, traceable evidence, and human oversight. FraudShield AI explores an evidence-first approach to financial fraud triage.

What it does

FraudShield AI analyzes three synthetic transaction scenarios using deterministic risk rules. It displays transparent risk scores, triggered indicators, SHA-256 evidence fingerprints, and downloadable JSON audit reports.

How we built it

We developed the prototype using Python, Streamlit, JSON, hashlib, and GitHub. The application also includes a configurable NVIDIA Nemotron integration using the Nebius Token Factory chat API.

Challenges

The main challenge is activating Nebius Token Factory because payment-card verification blocks access to promotional credits. Support ticket AISTUDIOSUP-2540 remains open. Live model inference has not yet been verified.

Accomplishments

We built and deployed an interactive prototype, tested synthetic risk scoring, and independently verified case and report SHA-256 values.

What we learned

Reliable AI investigation must distinguish evidence, heuristic indicators, AI interpretation, and human decisions.

What's next

Validate live Nemotron inference, expand automated tests, improve structured model output, and introduce richer evidence provenance and benchmarking.

Limitations

All included cases are synthetic. Risk scores are heuristics rather than fraud probabilities. The prototype does not establish wrongdoing, and cryptographic hashes are not digital signatures.piration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for FraudShield AI — Evidence-First Agent

Built With

• github?
• json?
• nebius?
• nemotron?
• nvidia?
• python?
• sha-256?
• streamlit?