Inspiration
Chargebacks and payment disputes create significant operational challenges for online merchants. Investigating a dispute often means manually searching invoices, tracking records, delivery confirmations, refund receipts, and customer communications under strict deadlines.
We identified another major problem: generic AI tools can generate convincing responses without verifying whether their claims are supported by actual evidence. In financial disputes, an invented tracking detail or unsupported statement can seriously damage a merchant's case.
We wanted to build something different: an AI-powered investigation platform where every important claim must be backed by documentary evidence.
What it does
ResolveIQ is an AI-powered PayPal Dispute Intelligence Platform that helps merchants investigate, evaluate, and respond to payment disputes.
The platform provides an end-to-end workflow:
- Dispute Ingestion: Connects with the PayPal REST API or uses clearly labelled demonstration cases to retrieve dispute information and response deadlines.
- AI Dispute Classification: Uses Google Gemini to understand customer complaints and classify disputes into relevant categories.
- Document Intelligence: Extracts information from PDF invoices, DOCX files, receipts, delivery records, and scanned documents using document parsers and OCR.
- Evidence Intelligence Matrix: Maps individual customer allegations against available documentary evidence and identifies supported, partially supported, and contradictory claims.
- Explainable Completeness Score: Calculates a 0–100 evidence completeness score based on documentary coverage and consistency, rather than inventing a probability of winning.
- Grounded Rebuttal Generation: Generates structured dispute responses where factual statements are linked to exact evidence excerpts.
- Human Approval: Requires explicit merchant verification and approval before evidence can be submitted.
- PayPal Integration: Supports structured evidence submission through the PayPal Disputes REST API.
- Zapier Automation: Sends dispute lifecycle notifications through configurable webhooks.
How we built it
We developed ResolveIQ using React 19, Vite, Tailwind CSS, Python, and FastAPI.
The frontend provides a merchant dashboard, dispute investigation workspace, evidence matrix, resolution studio, and automation settings.
The backend handles document processing, dispute classification, evidence matching, database operations, and external API integrations.
Google Gemini 2.5 Flash powers AI-based classification, document information extraction, and rebuttal generation. PyMuPDF and python-docx handle document parsing, while OCR provides an alternative for scanned documents.
SQLite provides persistent storage and audit logging. The PayPal REST API handles dispute-related operations, while Zapier webhooks support workflow automation.
We also implemented safeguards including prompt-injection isolation, file validation, API rate limiting, duplicate-submission protection, and PII masking.
Challenges we faced
One of our biggest challenges was ensuring that AI-generated responses remained grounded in actual evidence instead of producing plausible but unsupported statements.
We addressed this through evidence-to-claim matching, source-excerpt validation, missing-information detection, and explicit merchant-input requirements.
Another challenge was handling different document formats and inconsistent information across invoices, tracking records, and customer statements.
Integrating external payment APIs also required careful handling of authentication, token expiration, rate limits, and submission errors.
Finally, we designed a mandatory human approval mechanism to ensure that AI-generated responses are reviewed before submission.
What we learned
Building ResolveIQ taught us that useful AI systems need more than accurate language generation. They need traceability, validation, explainability, and appropriate human oversight.
We gained practical experience integrating Gemini AI, developing REST APIs, processing unstructured documents, designing evidence evaluation pipelines, and integrating payment workflows.
What's next
We plan to introduce distributed processing, PostgreSQL support, inbound PayPal webhooks, multi-tenant access control, and an advanced mobile document scanner.
Our long-term goal is to make dispute investigation faster, more transparent, and more reliable for businesses of all sizes.
Built With
- fastapi
- google-gemini-2.5-flash
- google-genai-sdk
- javascript
- paypal-ai-toolkit
- paypal-rest-api
- pillow
- pymupdf
- pytest
- python
- python-docx
- react-19
- rest-api
- sqlite
- tailwind-css
- uvicorn
- vite
- zapier-webhooks
