Inspir# ProofPulse AI — When a promise breaks, trust shouldn't.
Inspiration
Digital services are built on promises: availability, reliability, and trust. But when a paid service goes down, customers often face confusing explanations, slow support, and uncertain compensation. Small digital-service businesses face the opposite problem: they need a responsible way to investigate incidents and resolve them without manually coordinating every step.
We built ProofPulse AI to explore a simple idea: what if service recovery were transparent, evidence-grounded, and connected directly to the payment system?
What it does
ProofPulse AI is an experimental service-recovery studio for digital-service operators.
It brings together five stages:
- Payment: A PayPal Sandbox checkout workflow creates an order, supports buyer approval, and checks payment capture details.
- Detection: A monitor records actual HTTP responses and timestamps from a controlled demonstration service.
- Intelligence: A GroqCloud-hosted AI model interprets the recorded evidence and prepares a plain-language incident explanation.
- Decision: A deterministic policy checks eligibility for a sample $5 goodwill credit. The AI cannot change the amount or authorize money movement.
- Resolution: With explicit merchant authorization, the application supports a PayPal Sandbox partial-refund request and refund-status verification.
A Trust Ledger allows users to inspect an incident's evidence and export a privacy-conscious report.
How we built it
We used Python and FastAPI for the backend, SQLite for prototype persistence, and HTML, CSS, and JavaScript for the frontend. PayPal Orders v2 and Payments v2 power the sandbox payment workflow. GroqCloud serves the openai/gpt-oss-20b model for evidence-grounded incident analysis.
The interface uses an original dark editorial design, animated signals, contextual status indicators, and an interactive recovery experience.
We also implemented server-side financial rules, post-payment incident eligibility checks, explicit merchant authorization, and defenses against unsupported refund states.
Challenges we ran into
The biggest challenge was designing an AI experience around financial actions without allowing the AI to invent facts or make financial decisions.
We separated observed service data from model-generated explanations, and financial calculations from AI recommendations.
We also encountered a Windows HTTPS certificate-chain issue while integrating GroqCloud. We resolved it by using the operating system's trusted certificate store instead of disabling TLS verification.
Accomplishments
- Built an interactive service-monitoring and recovery prototype.
- Implemented a controlled endpoint that can produce observable HTTP 200 and 503 responses.
- Connected to the live GroqCloud model and verified a successful model response.
- Authenticated successfully with PayPal Sandbox OAuth.
- Implemented PayPal Sandbox order, capture, partial-refund, and refund-verification API workflows.
- Created an exportable incident evidence report with masked payment identifiers.
- Passed 15 automated tests, including mocked payment and AI integration tests.
What we learned
Trustworthy financial AI needs boundaries. AI is useful for understanding and communicating an incident, but it should not independently determine whether a failure occurred, invent compensation amounts, or claim a refund succeeded.
Those decisions require observable evidence, explicit rules, authorization, and confirmation from the payment provider.
Limitations
This hackathon prototype monitors its own controlled test service; it does not independently validate the availability of third-party platforms. Its $5 compensation rule is an illustrative policy, not a contractual service-level agreement.
At the time of writing, live GroqCloud model access and PayPal Sandbox authentication have been confirmed. A complete buyer-approved payment and refund cycle has not yet been independently verified. The project requires additional access controls and durable storage before production use.
What's next
We want to expand ProofPulse toward configurable service guarantees, verified monitoring integrations, merchant policy management, and privacy-conscious customer recovery workflows.
Our longer-term vision is to help small digital businesses make service recovery understandable, consistent, and accountable.ation
Built With
- css3
- fastapi
- gpt-oss-20b
- groqcloud
- html5
- javascript
- paypal-orders-api-v2
- paypal-payments-api-v2
- paypal-sandbox
- python
- sqlite
