ProofPay AI — Evidence First, Payment Second

Local AI reviews milestone delivery evidence before payment. Humans make the final decision, and PayPal Sandbox handles secure test checkout.

Comment

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for ProofPay AI — Evidence First, Payment Second

Inspiration

Millions of freelancers, agencies, and businesses face the same problem: how can a customer confidently pay for work when the quality and completeness of the delivery are difficult to verify?

Traditional payment systems process transactions, but they do not assess whether an agreed milestone has actually been delivered.

ProofPay AI introduces a different approach: Evidence First. Payment Second.

Our vision is to make milestone payments more transparent, explainable, and trustworthy by combining local AI, structured evidence checks, human decision-making, and PayPal.

What it does

ProofPay AI is an open-source prototype that helps buyers evaluate evidence of completed work before making a payment.

The workflow is straightforward:

  1. A project milestone defines the expected deliverables.
  2. The seller provides delivery evidence.
  3. Deterministic checks and a local AI model analyze the submitted evidence.
  4. The system presents an advisory assessment with understandable reasons.
  5. The buyer reviews the results and explicitly approves or rejects the milestone.
  6. Only after human approval can the application initiate a PayPal Sandbox checkout.

The AI never has the authority to approve or release a payment by itself.

A separate interactive visualization, called Evidence Constellation, helps demonstrate how project requirements, submitted evidence, AI uncertainty, human approval, and payment status are connected.

How we built it

ProofPay AI was developed using Node.js, JavaScript, HTML, CSS, Transformers.js, a locally executed MiniLM/ONNX embedding model, and PayPal Orders v2.

The system combines two complementary approaches:

  • Deterministic validation: transparent evidence checks that produce understandable audit reasons.
  • Local AI inference: semantic similarity analysis that helps evaluate whether submitted evidence relates to the agreed milestone.

AI recommendations remain advisory. Explicit human approval is required before checkout.

PayPal Developer Sandbox provides realistic payment integration without transferring real money.

On October 27, 2026, the application-level workflow was tested against PayPal Sandbox, including buyer authorization, payment capture, and independent confirmation that the test transaction reached COMPLETED status.

Automated local tests also cover core application behavior and payment-related logic.

Challenges we faced

One major challenge was designing an AI-assisted payment process without giving AI unrestricted control over financial decisions.

Another challenge was distinguishing between meaningful evidence and misleading or incomplete submissions.

Semantic similarity alone cannot establish that work is authentic, complete, or high-quality. Therefore, ProofPay AI combines AI suggestions with deterministic checks and mandatory human review.

Integrating the PayPal Sandbox order lifecycle, buyer approval, and transaction verification also required careful separation of simulated demonstrations from real Sandbox API interactions.

Accomplishments

We created a working local prototype that demonstrates:

  • Local AI inference without requiring a hosted AI inference service.
  • Explainable milestone evidence checks.
  • Human-controlled payment authorization.
  • PayPal Orders v2 Sandbox integration with a verified completed test capture.
  • An interactive Evidence Constellation visualization.
  • Automated tests and publicly available MIT-licensed source code.

Our demonstration video distinguishes simulated workflow elements from actual PayPal Sandbox interactions.

What we learned

AI can support financial decisions without becoming the final decision-maker.

A trustworthy payment workflow requires more than a model-generated score. It needs transparent reasons, controlled authorization, clear limitations, and verifiable payment states.

We also learned that AI uncertainty should be communicated openly rather than hidden behind an apparently precise confidence score.

What's next

Our next development priorities include:

  • Stronger evidence authenticity verification.
  • Secure user authentication and multi-tenant support.
  • More advanced AI-assisted document analysis.
  • Improved transaction event reconciliation.
  • Better audit trails and dispute-support capabilities.
  • Moving from a local prototype toward a securely hosted application.

ProofPay AI is not yet a production payment service, escrow platform, or automated proof-of-delivery system.

Our long-term vision is simple: make digital payments more trustworthy by evaluating delivery evidence before payment — while keeping humans in control.

Evidence First. Payment Second.

Built With

Share this project:

Updates

Submission history