PayPal Fraud & Dispute Triage Agent

Inspiration

High-volume payment platforms face immense operational friction processing transaction disputes manually, causing support teams to waste valuable hours reviewing straightforward low-risk cases instead of resolving critical issues.

What it does

An autonomous AI triage agent that ingests incoming transaction dispute payloads, evaluates risk factors, and instantly categorizes cases into auto_resolve, escalate_fraud, or needs_review, returning strict JSON outputs.

How we built it

Built using Relevance AI to configure the core system prompts, execution logic, and structured JSON schema validation layers for seamless downstream data routing.

Challenges we ran into

Balancing flexibility with strict schema enforcement to ensure the agent never breaks conversational character and always returns clean, machine-readable JSON actions.

Accomplishments that we're proud of

Successfully deploying a fully functional low-code triage engine that handles risk assessment and automated decision-making without manual intervention.

What we learned

Rigorous prompt structuring and explicit schema definitions are critical when building reliable autonomous agents for financial workflows.

What's next for PayPal Fraud & Dispute Triage Agent

Integrating live webhook endpoints and expanding downstream action workflows to execute automated refunds and security escalations natively.

Built With

• artificialintelligence
• json
• no-code
• relevance-ai