Software Engineer · AI Code Evaluation & Model Training
- 50/50
- How my week splits
- 3
- AI training platforms
- 3 yrs
- Building products
Shipping product code · Training models
Mercor · Alignerr · AfterQuery
React · Next.js · TypeScript · Web3
About
CurrentlySelf-employed, taking contract AI evaluation work and selective freelance builds.
AI Code Evaluation & Model Training
The work splits into two kinds of task. Both come down to reading a large unfamiliar codebase quickly and telling a correct fix from one that merely looks correct.
Feedback & Evaluation
- Implement real features and fixes on large open-source repositories, then use that work as ground truth for model attempts at the same problems.
- Review model-generated code on correctness, approach, and instruction following, not just whether it runs.
- Assess web page builds and UI code where the failure is visual or structural rather than a thrown error.
Adversarial Task Authoring
- Write the instructions, reference solution, and test suite for problems built to defeat frontier models.
- Validate every task by running the model against it. A task only counts when the model fails.
- Target failure modes that make a problem genuinely hard rather than ambiguous, so a failure points at a real capability limit.
Agent Transcript Review
- Trace where the model's reasoning breaks down instead of only checking whether the output passes.
- Run side by side evaluations of two models on one task, assessing each independently before comparing.
- Judge whether the agent recovered from its own errors or kept compounding them.
Written Rationales
- Explain why a solution succeeds or fails at a level another engineer can act on without rereading the diff.
- Keep judgments consistent across a batch so the training signal stays clean.
- Turn isolated mistakes into named, recurring failure modes a team can actually work against.
Task types I evaluate
Work Experience
AI Evaluation & Model Training
Product Engineering
Skills
AI Evaluation & Model Training
AI Tooling
Languages
UI / Frontend
State & Data
Backend & DB
Web3 / Blockchain
Testing
Tools & Infra
Auth & APIs
Check out my latest work
AI products, full-stack apps, and on-chain systems I built end to end. This is the engineering background that makes my model evaluations worth reading.
Resume Genie(AI Resume Builder)
An innovative full-stack web application that helps users create professional resumes with AI-powered generation. Features seamless customization, real-time preview, secure Clerk authentication, and the ability to save and share resumes via unique links. Powered by Gemini API for intelligent resume optimization, ensuring ATS-friendly and industry-standard formatting. Users can generate, customize, and share their resumes with potential employers through an intuitive interface with live preview updates. The platform integrates MongoDB for data storage, Node.js backend, and TailwindCSS for responsive design across all devices. Future enhancements include template customization, LinkedIn integration, PDF export, and AI-powered cover letter generation.
EtherGuild(Decentralized Freelancing Platform)
A revolutionary decentralized freelancing platform combining blockchain technology with modern web development. Built on Ethereum, it enables secure, transparent, and trustless interactions between clients and freelancers worldwide. Features Web3 integration with MetaMask and WalletConnect, smart contract-based escrow payments, on-chain reputation system, and a modern responsive UI built with React and Tailwind CSS.
Formify AI(AI Form Builder)
A powerful and user-friendly application for creating customized forms effortlessly using AI. Powered by Gemini AI API, users can generate form fields based on prompts with full customization capabilities. Features include secure Next-auth authentication, responsive design with Shadcn UI, persistent PostgreSQL data storage, and admin dashboard for managing submissions. Users can create any form type, share via generated links, customize themes from six options, export form data to Excel, and control authentication requirements. Supports anonymous or authenticated form submissions with complete form data management. Built with Next.js for SSR, Drizzle ORM for database interactions, and DaisyUI for styling.
Admin Dashboard
A React-based CRM dashboard featuring robust authentication, real-time data management, and an interactive Kanban board. Leverages Refine, GraphQL, Ant Design, and Vite for fast, responsive admin interface. Includes secure login/signup, granular access control authorization, dynamic dashboards with interactive charts, full CRUD functionality for companies and sales management, real-time Kanban board with due dates and multi-assignees, user profile management, and fully responsive design. Features real-time activity updates, sales analytics, task management with markdown descriptions, company profiles with search/pagination, and personalized account settings across all devices.
Let's build better models
Self-employed, taking contract AI evaluation work and selective freelance builds. Pick whichever way is easiest.
Email me
Usually replies within a few hours
Book a 15 minute call
Straight into my calendar