Inspiration
The inspiration behind ProofLens came from an incident involving a resin artist who invested significant time, effort, and materials into creating a customized product, only to have the customer refuse payment because the finished product did not meet their expectations.
This made me think: What happens when customers and service providers have different understandings of what was promised and delivered?
The problem extends beyond handmade products. It exists in custom jewellery, furniture, property renovation, maintenance, website development, and freelance services.
Customers need a way to establish what they requested, while service providers need a way to demonstrate what they delivered.
We wanted to build a platform that could bridge this trust gap by documenting agreements, tracking progress, and connecting evidence with payment events. That idea became ProofLens.
What it does
ProofLens – Multimodal Escrow and Dispute Copilot is an AI-powered agreement-to-evidence platform that helps customers and service providers establish a transparent record throughout a project.
It supports both physical and digital projects.
Key capabilities include:
- AI-Powered Agreement Creation: Converts natural-language conversations into structured agreements covering scope, materials, measurements, milestones, deadlines, pricing, and acceptance criteria.
- Baseline Documentation: Records initial conditions and agreed requirements before work begins.
- Multimodal Evidence Capture: Allows service providers to submit photographs through live camera capture or upload relevant project files.
- Context-Aware AI Assessment: Analyzes submitted images and project context against agreed requirements, identifying visible inconsistencies, missing evidence, and areas requiring further review.
- Measurement Validation: Compares submitted measurement values against agreed dimensions and tolerances using deterministic backend validation.
- Scope Change Tracking: Documents modifications and customer-approved changes throughout the project.
- Evidence Timeline: Maintains a traceable history of agreements, submissions, observations, approvals, and project events.
- Evidence Reports: Consolidates requirements, submitted evidence, assessment results, and unresolved issues into a structured report.
- PayPal Integration: Connects project milestones with actual PayPal payment events, providing greater visibility into the relationship between project progress, customer decisions, and payments.
ProofLens does not simply approve or reject work. It helps both parties understand what the evidence supports and what still requires attention.
How we built it
We built ProofLens around a shared project lifecycle involving customers and service providers.
The application brings together frontend interfaces, backend services, multimodal AI, evidence management, and PayPal payment integration.
The core workflow is:
- The customer creates a project and describes their requirements.
- AI converts those requirements into a structured agreement.
- Both parties review and confirm the agreement.
- The initial condition is documented through baseline evidence.
- The service provider submits progress evidence throughout the project.
- Multimodal AI evaluates relevant evidence against the agreement.
- Backend validation checks numerical measurements and tolerances.
- Approved scope changes and customer decisions are recorded.
- PayPal payment events are connected to the project journey.
- A consolidated evidence report provides a transparent project history.
We designed the platform to support different project categories using the same underlying workflow, including custom physical products, property maintenance, and digital deliverables.
Challenges we ran into
1. Understanding context rather than just images
Analyzing an image independently is insufficient. The AI needs to understand the original agreement, baseline condition, relevant requirements, and previous updates before assessing submitted evidence.
2. Physical measurement verification
A photograph alone cannot guarantee accurate dimensions. We addressed this by separating AI-assisted measurement extraction from deterministic backend comparisons against agreed tolerances.
3. Evidence authenticity
A live camera capture or image hash does not guarantee that a photograph represents genuine work. We had to consider duplicate images, screen recaptures, manipulated-looking images, unclear photographs, and insufficient evidence while avoiding unsupported claims of fraud detection.
4. PayPal payment integration
Connecting project milestones with payment operations required careful handling of payment authorization, capture, transaction status, and the actual capabilities of PayPal APIs. We focused on accurately representing payment events rather than claiming unsupported escrow functionality.
5. Building a complete, responsive experience
Bringing together customer and service-provider workflows, mobile evidence capture, AI assessments, project tracking, and payment integration while keeping the application intuitive was another major challenge.
Accomplishments that we're proud of
- Built a unified platform designed for both physical and digital projects.
- Developed an AI-assisted workflow that transforms informal requirements into structured agreements.
- Integrated contextual multimodal evidence assessment into the project lifecycle.
- Designed measurement validation that combines AI assistance with deterministic backend logic.
- Created a traceable project timeline covering agreements, evidence, changes, and approvals.
- Integrated PayPal payment workflows into the project experience.
- Designed responsive experiences for customers and service providers, including mobile evidence capture.
- Focused on transparent AI assessments that can identify uncertainty instead of automatically declaring every submission successful.
Most importantly, we are proud of building ProofLens around fairness for both sides, rather than treating dispute resolution as a one-sided approval process.
What we learned
Building ProofLens taught us that meaningful AI integration involves much more than connecting an API to an application.
We learned how to:
- Use multimodal AI to interpret images alongside structured project context.
- Convert unstructured conversations into clear, versioned requirements.
- Combine AI reasoning with deterministic backend validation.
- Design evidence workflows that preserve traceability.
- Handle uncertainty and distinguish AI observations from verified facts.
- Integrate payment operations while respecting actual API capabilities.
- Think about trust, transparency, and fairness as core product requirements.
One of our biggest takeaways was that a trustworthy AI system should be able to say, "There isn't enough evidence to verify this yet," rather than confidently generating an unsupported conclusion.
What's next for ProofLens - Multimodal Escrow and Dispute Copilot
We envision ProofLens evolving into a broader trust and evidence infrastructure for service-based transactions.
Our next steps include:
- Improving multimodal evidence analysis and contextual reasoning.
- Exploring stronger evidence provenance and authenticity checks.
- Supporting specialized measurement and independent inspection integrations.
- Expanding support for digital deliverables, including website and software development.
- Enhancing dispute resolution with structured evidence comparisons and neutral summaries.
- Exploring additional PayPal-supported payment workflows for milestone-based transactions.
- Extending the platform to property managers, Airbnb owners, hotels, agencies, and customers managing projects remotely.
Our long-term vision is to make every project more transparent, every agreement clearer, and every disagreement easier to understand.
ProofLens: From Agreement to Evidence. From Uncertainty to Trust.
Built With
- ag-grid
- angular-material
- angular.js
- fastapi
- google-gemini
- html5
- httpx
- jwt
- node.js
- paypal
- paypal-disputes-api
- paypal-invoicing-api
- paypal-oauth
- pillow
- postgresql
- pydantic
- python
- render
- rxjs
- scss
- sqlalchemy
- sqlite
- typescript
- uvicorn
- vercel