Inspiration

Enterprise commerce moves fast, but traditional B2B settlement mechanisms create severe operational bottlenecks. Manual approvals slow down supply chains, while unguided automation opens the door to fraud, compliance failures, and financial disputes. We built Afferent Commerce to bridge the gap between physical quality telemetry and automated financial execution—ensuring enterprise trust at scale.

What it does

Afferent Commerce is an autonomous agentic commerce engine that links real-time physical product quality signals and somatosensory telemetry to automated financial settlement. Guided by the core principle: "AI recommends. Human approves. PayPal executes.", the platform:

  1. Ingests Physical Telemetry: Captures acoustic friction, surface deformation, and structural quality signals.
  2. AI Risk & Quality Assessment: Leverages NVIDIA Nemotron to generate real-time risk scores (0-100) and actionable recommendations (APPROVE, REVIEW, HOLD, REJECT).
  3. Human-in-the-Loop Governance: Enforces explicit human operator authorization before financial movement.
  4. Automated PayPal Execution: Executes real-time mass settlements via PayPal Payouts REST API v1.
  5. Immutable Audit Trail: Logs every decision, risk metric, and PayPal transaction identifier into an auditable ledger.

How we built it

  • AI Engine: NVIDIA Nemotron for deep risk evaluation and multi-tier merchant diagnostic scoring.
  • Financial Execution: Integrated PayPal Payouts REST API v1 utilizing OAuth2 Client Credentials for real-time batch settlement and item status tracking.
  • Backend Infrastructure: Built with FastAPI (Python 3.11), Pydantic, Uvicorn, and Docker Compose for cloud-native scalability.
  • Command Center Dashboard: Designed with React, Vite, TypeScript, Tailwind CSS, and Framer Motion following an executive 3-6-3 grid layout with real-time system health observability.

Challenges we ran into

Integrating non-deterministic LLM reasoning (NVIDIA Nemotron) with deterministic financial settlement APIs required strict boundary enforcement. We solved this by implementing deterministic schemas, confidence threshold guards, and an explicit Human-in-the-Loop circuit breaker to ensure no capital moves without explicit human validation.

Accomplishments that we're proud of

  • Achieving end-to-end execution from raw tactile telemetry down to verified PayPal Sandbox transaction identifiers.
  • Designing a zero-friction UI that presents 99% confidence scores and real-time audit trails for financial compliance teams.
  • Delivering a production-ready, Dockerized architecture with 100% clean test suite passes.

What's next for Afferent Commerce

  • Expanding multi-sensor physical telemetry support (thermal, hyperspectral, and visual defect inspection).
  • Multi-currency dynamic riskhedging natively inside the PayPal Payouts engine.
  • Enterprise ERP connectors (SAP S/4HANA and Oracle Cloud SCM).

Built With

Share this project:

Updates

posted an update —

Afferent Commerce is officially submitted to the PayPal AI Hackathon!

Hello everyone! We are thrilled to announce that Afferent Commerce has been officially submitted to the PayPal AI Hackathon.

We have successfully completed the end-to-end architecture bridging high-precision physical telemetry, autonomous risk reasoning via NVIDIA Nemotron, explicit human-in-the-loop governance, and real-time financial execution through the PayPal Payouts v1 API.

You can check out the full project story, open-source codebase, and technical demo video right here on Devpost and via our official GitHub repository. We continue pushing the boundaries between AI and autonomous enterprise commerce!

Submission history