VERIFAI

AI copilot that triages PayPal disputes, auto-assembles carrier evidence, and drafts winnable rebuttals for 1-click merchant submission.

Comment

Inspiration

When a customer opens a payment dispute on PayPal, the merchant has only a few days to find tracking numbers, postal receipts, and chat history to prove they did their job. If they miss the deadline, PayPal automatically refunds the buyer. The seller loses the item, loses the money, and gets hit with an extra $15 dispute fee.

Big companies have entire legal and fraud teams for this. Small business owners just have late nights. We built VERIFAI to help independent sellers defend themselves against unfair chargebacks in seconds.

What it does

VERIFAI is an AI copilot that protects merchants against payment disputes:

  • Gathers All the Proof: Connects to the PayPal Customer Disputes API v1 and automatically matches active disputes with store orders, carrier tracking status, and buyer chat logs into an indexed Evidence Pack (F1, F2, F3...).
  • Deterministic Winnability Scoring (0–100): Evaluates dispute defensibility using a mathematical formula based on PayPal Seller Protection rules:

$$ \text{Score} = \sum w_i \cdot \text{evidence}_i $$

It recommends the winning operational strategy: Fight (70+), Settle (40–69), or Accept (under 40).

  • Sub-Second AI Defense Drafting: Uses Groq AI (openai/gpt-oss-120b) to write a formal, persuasive response to PayPal in under 1 second, citing verified proof like [F1] or [F2].
  • Human-in-the-Loop Safety: An automated citation validator catches hallucinations before review. The AI never submits claims or refunds without the merchant clicking "Approve".
  • Financial Analytics: Features an AG Grid dashboard tracking frozen dispute funds, fees saved, and historical win rates.

How we built it

  • Frontend: Built with React 18, TypeScript, and Tailwind CSS, using AG Grid Community for high-performance financial data tables and interactive order inspection drawers.
  • Backend: Built with Node.js and Express (TypeScript), with strict Zod schema validation and full OpenAPI 3.0 documentation.
  • AI Engine: Powered by Groq for high-speed inference, backed by an automated fact-checker and an offline Grounded Rule Engine fallback if cloud services are unavailable.
  • Developer Tools: Pre-generated client SDKs for Python and TypeScript so merchants can integrate VERIFAI into their checkout platforms in 3 lines of code.

Challenges we ran into

You cannot let AI guess when real money is on the line.

In early tests, standard AI prompts would sometimes hallucinate: inventing delivery dates, assuming the buyer signed for a package, or imagining customer admissions. In payments, submitting false evidence can get a merchant banned for fraud.

We solved this with a strict separation of duties:

  1. Math calculates the score: An objective formula handles 100% of the winnability score with zero AI hallucinations.
  2. AI is locked to evidence: The model is restricted to facts in the Evidence Pack and must cite each one.
  3. Automated Citation Checker: If the AI cites a fact that doesn't exist, our validator rejects the draft and forces an instant rewrite before the merchant ever sees it.

Accomplishments that we're proud of

  • Sub-Second Speed: Generates complete, evidence-backed defense letters in under 800ms using Groq.
  • Deterministic Grounded Scoring: Backed by 23 passing unit tests in Vitest covering edge cases like late deliveries and missing signatures.
  • Financial Safeguards: Built-in simulation mode (DRY_RUN=true) and automatic currency checks so settlement offers match the original transaction currency without overpaying.

What we learned

Trust in AI in finance is not created by giving an autonomous agent a blank check to move money. It is built through transparency, speed, and boundaries: giving the merchant clear mathematical odds, verified citations, and the final decision button. Turning a 45-minute administrative headache into a 30-second review changes the economics of fighting friendly fraud.

What's next for VERIFAI

  • Live Carrier Webhooks: Direct tracking event updates from USPS, FedEx, and UPS.
  • Platform Integrations: One-click apps for Shopify, WooCommerce, and BigCommerce stores. -Pre-Dispute Mediation and simultaneous evidence collection: Analyzing early customer chat sentiment to resolve buyer friction before it escalates into a formal chargeback; like collecting and managing evidence simultaneously, autonomously.

Built With

Share this project:

Updates

Submission history