Inspiration
Freelancing and software development are built on trust, but payments often create uncertainty for both clients and developers. Clients hesitate to pay upfront because they cannot always verify the quality of the deliverables. Freelancers, on the other hand, risk completing hours of work without receiving timely payment.
Traditional payment workflows do not always provide a fair, transparent way to handle incomplete work, milestone disagreements, or payment disputes.
We wanted to solve this problem by building something beyond a traditional payment platform: an intelligent system that evaluates work, manages milestone-based payments, and makes the settlement process more transparent.
That idea led to Lancer AI, an autonomous escrow-style platform powered by AI and PayPal.
What it does
Lancer AI connects clients and freelancers through a milestone-based contract and payment system.
Instead of releasing the entire payment upfront, clients authorize funds through PayPal. Freelancers submit their completed work along with supporting evidence, such as GitHub pull requests, live staging URLs, Figma designs, and screenshots.
The platform then uses a 3-Agent AI Tribunal to evaluate the submission:
Agent 1: Spec & Code Auditor Examines GitHub pull requests, code changes, acceptance criteria, security concerns, and technical evidence.
Agent 2: Multimodal Vision & UX Inspector Evaluates visual quality, responsiveness, design consistency, and user experience using staging information and visual evidence.
Agent 3: Financial Risk Arbiter Combines evaluation scores, assesses the payment risk, and determines the appropriate milestone settlement. Based on the final evaluation, Lancer AI supports three outcomes:
Full Approval: The milestone is approved and the eligible payment is released.
Partial Approval: A percentage of the payment is released while the remaining amount is held for additional work.
Changes Requested: Payment is not released, and the freelancer receives a remediation checklist.
This makes payment decisions more structured, transparent, and evidence-based.
How we built it
We built Lancer AI using Next.js, React, and TypeScript, with a backend architecture designed around AI evaluation and payment workflows.
The major technologies and integrations include:
Google Gemini: Powers AI-based code evaluation, reasoning, and multimodal inspection.
PayPal Orders v2 API: Creates payment orders using authorization rather than immediate capture.
PayPal Captures v2 API: Supports milestone-based partial payment releases.
PayPal Payouts API: Supports splitting milestone payments among multiple freelancers or agency members.
GitHub REST API: Retrieves pull request information and changed-file telemetry.
SHA-256: Generates cryptographic verification certificates linked to milestone evidence.
Supabase: Provides database integration capabilities.
Next.js API Routes: Handle application logic, AI evaluation, payment operations, and webhook workflows.
We also designed a contract-generation workflow that converts plain-language project requirements into structured milestones and acceptance criteria.
For the demonstration, we created three evaluation scenarios to show how the tribunal responds to successful, partially successful, and rejected submissions. The platform also supports live Gemini evaluation and external evidence inspection when the required credentials and inputs are configured.
Challenges we ran into
One of the biggest challenges was designing an AI evaluation system that could assess different types of deliverables rather than relying on a single generic AI response.
Coordinating three specialized agents required consistent evaluation outputs, structured JSON responses, and a reliable scoring mechanism.
Another major challenge was integrating payment authorization and partial capture. We had to carefully handle payment states so that approving one milestone would not unintentionally release the remaining contract funds.
We also worked on handling API failures, validating external URLs and evidence, managing webhook verification, and maintaining a clear audit trail for payment decisions.
Balancing AI flexibility with predictable financial decisions was particularly challenging. We addressed this by combining AI-generated evaluations with structured scoring, explicit settlement rules, and deterministic demonstration scenarios.
Accomplishments that we're proud of
Building Lancer AI taught us how to combine multiple AI agents with real-world financial workflows.
We gained practical experience with:
Building multi-agent AI architectures.
Integrating Google Gemini for text and visual analysis.
Working with PayPal authorization, partial capture, and payout APIs.
Retrieving and evaluating GitHub development evidence.
Designing milestone-based financial workflows.
Handling asynchronous API operations, validation, and webhook security.
Generating cryptographic hashes for tamper-evident records.
Most importantly, we learned that building an AI-powered financial platform requires more than generating intelligent responses. It requires reliable integrations, transparent decisions, careful validation, and a system that users can understand and trust.
What we learned
Building Lancer AI taught us how to combine multiple AI agents with real-world financial workflows.
We gained practical experience with:
Building multi-agent AI architectures.
Integrating Google Gemini for text and visual analysis.
Working with PayPal authorization, partial capture, and payout APIs.
Retrieving and evaluating GitHub development evidence.
Designing milestone-based financial workflows.
Handling asynchronous API operations, validation, and webhook security.
Generating cryptographic hashes for tamper-evident records.
Most importantly, we learned that building an AI-powered financial platform requires more than generating intelligent responses. It requires reliable integrations, transparent decisions, careful validation, and a system that users can understand and trust.
What's next for Lancer AI
Our vision is to evolve Lancer AI into a complete trust and payment infrastructure for the growing freelance and AI-agent economy.
Future improvements include more advanced code and visual verification, broader project-management integrations, stronger dispute-resolution workflows, and expanded support for distributed development teams.
Lancer AI is our step toward making freelance payments smarter, fairer, and more transparent, where completed work can be verified before payments are released.
Built With
- api
- gemini
- next.js
- react
- supabase
- tailwind
- typescript