Back to projects

Explain My Contract (AI-powered Legal Document Explainer)

Designed, built, and deployed a production-ready AI web application that simplifies complex contracts for users, with a strong focus on safety, observability, reliability, and controlled usage.

Impact

Shipped a production-ready AI product rather than a prototype.
Demonstrated full ownership from idea → build → deploy → monitor.
Applied DevOps thinking to a personal project, not just infrastructure work.

I wanted to build a real-world product that goes beyond a demo app, focusing on production concerns like abuse prevention, monitoring, safe AI usage, and operational visibility.

  • Contracts are difficult to understand for non-legal users, especially freelancers and contractors.
  • Generic AI tools don't provide structured, safe, or purpose-built explanations.
  • AI-powered apps are easy to misuse or abuse if usage limits and state handling aren't designed properly.
  • Many personal projects ignore monitoring, alerting, and reliability entirely.
  • Build a focused AI product with a clear, narrow use case rather than a generic chatbot.
  • Ensure explanations are structured, cautious, and clearly not legal advice.
  • Design the system to prevent abuse and unintended repeated usage.
  • Treat the project like a production service with monitoring, health checks, and alerts.
  • Ship quickly while maintaining operational discipline.
  • Next.js application using the App Router for clean routing and server/client separation.
  • Server-side API routes handle document processing and AI requests.
  • Explicit session and state management to control usage and prevent repeated generation abuse.
  • Stateless frontend with controlled backend logic to enforce limits and flow.
  • Deployed on Vercel with environment-based configuration (no secrets in code).
  • Git-based workflow: feature branches → PR → main.
  • Automatic builds and deployments triggered by GitHub pushes via Vercel.
  • Production deployments are deterministic and repeatable.
  • Environment variables managed securely by the hosting platform.
Purpose-built AI flow instead of a generic chat interface
Less flexible than free-form chat, but far safer, clearer for users, and easier to control.
Strict session and navigation control
More upfront logic, but prevents abuse such as repeated regeneration via browser navigation.
Production monitoring from day one
Extra setup effort, but provides real insight into usage, failures, and availability.
  • No secrets stored in the repository or client-side code.
  • All sensitive configuration handled via environment variables.
  • Clear legal disclaimers to prevent misuse or reliance as legal advice.
  • Input handling designed to avoid accidental data retention.
  • Vercel Analytics used to observe real-world usage patterns and performance.
  • Health check endpoints implemented to confirm application availability.
  • External uptime monitoring (UptimeRobot) with alerting to detect outages quickly.
  • Simple, predictable architecture reduces failure modes.
  • Ran as a live, publicly accessible application with controlled AI usage (since retired).
  • Clear operational visibility through analytics and uptime monitoring.
  • Identified and fixed a real abuse vector through testing and iteration.
  • Reusable foundation for future AI-powered tools or services.
  • Add rate-limiting at the API layer for additional abuse protection.
  • Introduce basic synthetic checks for deeper health validation.
  • Add structured logging for better debugging and post-incident analysis.
  • Optional PDF export and history handling with explicit user consent.