💡 Inspiration

Every year, online merchants lose over $100B to cart abandonment and merchant chargebacks. Small and medium businesses don't have dedicated 24/7 fraud ops teams to catch failed checkout sessions or assemble evidentiary packets within PayPal's strict dispute response deadlines.

We built Transaction-Autopilot to bridge this gap: an autonomous AI copilot and telemetry engine that monitors PayPal transaction lifecycles, automates dispute defense, and recovers lost revenue with zero human friction.


🚀 What It Does

Transaction-Autopilot provides an enterprise-grade command center for PayPal merchants:

  1. Orders v2 Telemetry & Forensics: Real-time visibility into PayPal Orders v2 lifecycles (APPROVAL_PENDING, COMPLETED, VOIDED), isolating stranded buyer checkouts and computing real-time recovery scores.
  2. Intelligent Dispute Defense: Automatically maps incoming claims (e.g. Unauthorized Transaction, Item Not Received), pairs them with verified carrier tracking, and compiles evidence packets ready for 1-click PayPal resolution.
  3. Autonomous AI Recovery Assistant: Powered by Llama 3 via Groq, the copilot analyzes merchant sales velocity, answers natural-language store queries, and dispatches compliant smart payment recovery links via Brevo SMTP.
  4. Cryptographic Audit Trail: Tamper-evident SHA-256 event ledger capturing every state mutation for strict compliance and auditability.

🛠️ How We Built It

  • Frontend: React 19, TypeScript, Vite, Tailwind CSS, Lucide icons, responsive high-contrast fintech design system.
  • Backend: Node.js, Express, RESTful architecture with atomic JSON persistence.
  • PayPal Integration: Modern PayPal REST APIs (/v2/checkout/orders), Sandbox OAuth 2.0 token handshakes, and smart button workflows.
  • AI & Messaging: Groq LPUs running Llama 3 (ultra-low latency inference) and Brevo Transactional SMTP for customer outreach.
  • Testing & CI/CD: Comprehensive 145-test automated test suite (OAuth, role separation, security invariants, SMTP validation) running on GitHub Actions CI/CD.

⚡ Challenges We Ran Into

  • State Machine Synchronization: Handling asynchronous order approval transitions without race conditions required strict idempotency keys.
  • Financial Determinism in AI: Ensuring the AI assistant never hallucinated monetary amounts or supported unverified currencies. We enforced strict USD ($) normalization across 100% of data structures.
  • Automated CI/CD in Cloud Runners: Decoupling live credential checks from unit test suites so GitHub Actions cloud runners verify builds seamlessly without leaking secrets.

🏆 Accomplishments That We're Proud Of

  • 145/145 Automated Tests Passing: 100% pass rate across security, OAuth verification, data persistence, and API boundaries.
  • Live Production Cloud Deployment: Fully operational at high availability on Render.
  • Open-Source Recognition: Selected by the XERJ Open-Source Corpus Hub as a reference architecture for secure agentic fintech implementations.

📚 What We Learned

  • Advanced PayPal Orders v2 token handshakes and dispute adjudication lifecycles.
  • Designing resilient prompt constraints for financial AI agents where precision is mission-critical.

🔮 What's Next for Transaction-Autopilot

  • Direct PayPal Disputes API v1 webhook listeners for fully hands-free proof-of-delivery dispatch.
  • Multi-merchant organization switcher with granular team role permissions.
  • Real-time predictive fraud anomaly scoring prior to order capture.

Built With

Share this project:

Updates

Submission history