Hackathon Project·May – Jul 2025·HackOn with Amazon, Season 5

Programmable Escrow & Fraud-Resilient Marketplace Payments

A programmable Solana escrow and event-driven fraud-risk architecture for marketplace settlement and post-return payment release.

Cover image for Programmable Escrow & Fraud-Resilient Marketplace Payments

Overview & Research Motivation

Built a marketplace payment architecture designed to reduce refund and settlement fraud by combining programmable on-chain escrow with event-driven risk scoring and settlement workflows.

The Core Systems Problem

  • Marketplace settlement must balance timely seller payment with the risk of refund abuse, return fraud, and premature release of funds after a return event.
  • Risk signals and transaction events need to move through a reliable workflow before programmable escrow releases or retains settlement funds.

My Specific Technical Contributions

  • Built a Solana escrow protocol in Rust and Anchor using PDA-controlled vaults and programmable settlement logic to automate post-return fund release and reduce refund and settlement fraud.
  • Engineered an AWS event-driven backend using EventBridge, Lambda, Step Functions, and DynamoDB.
  • Integrated Amazon Fraud Detector for transaction risk scoring.

System Architecture & Verification Pipeline

Escrow Architecture

Solana programs written with Rust and Anchor use Program Derived Address-controlled vaults and programmable settlement rules to hold and release marketplace payments.

Fraud-Risk Pipeline

Transaction and return events are evaluated through Amazon Fraud Detector before the settlement workflow reaches a release decision.

Event-Driven Backend

AWS EventBridge routes events into Lambda functions and Step Functions workflows, with DynamoDB retaining the state needed for settlement processing.

Settlement Workflow

Risk-scoring outputs and post-return events inform programmable keep-or-release transitions for funds held in escrow.

Evaluation & Benchmark Results

50K+
Transactions/day
Supplied system-scale figure
95%
Risk-scoring accuracy
Supplied evaluation figure

Technical Stack & Tools

RustAnchorSolanaProgram Derived AddressesAWS EventBridgeAWS LambdaAWS Step FunctionsDynamoDBAmazon Fraud Detector

Known Limitations

  • The prototype was developed for HackOn with Amazon, Season 5 and was not presented as a production deployment.
  • Risk-scoring quality depends on representative transaction and return-event data.

What I Would Test Next

  • Evaluate settlement behavior under a wider range of marketplace return and dispute scenarios.
  • Test failure recovery and idempotency across the event-driven settlement workflow.

Connected Systems & Inquiries