The challenge
Taskimon pairs a task manager with RPG progression, so finishing real work levels up your character. The tricky part is the reward. Every task needs a fair EXP value, and players can’t be trusted to rate their own tasks. So an AI evaluator scores each task instead, and it has to give consistent results at a cost that stays under control.
My role
I built Taskimon on my own as the final thesis project for my BSc (Hons) in Applied Computer Engineering at Goldsmiths, University of London and Algebra University: the Unity client, the backend and the AI integration.
The backend is a layered ASP.NET Core Web API using CQRS over PostgreSQL on Supabase. The app also uses LootLocker as a game-service backend, so I designed a split-backend security model that maps one verified identity across both.
Whenever a task is created or edited, the OpenAI API returns a time estimate, its reasoning and an EXP reward as structured output. A deterministic mapping turns that into the final reward, so the same task always earns the same amount. Re-evaluation rules, response caching and rate-limit handling keep the results consistent and the costs low.
Around that sits the game layer: EXP, levels, tiers, currency, a cosmetics shop, character customization, and friends whose profiles you can inspect.
Results
Taskimon runs on iOS and Android but isn’t publicly released. It’s full-stack work from end to end: a Unity client, a custom .NET backend, a third-party game backend and an LLM integration designed to be predictable rather than random. It uses the same architecture I use on client work.
Tech
Unity, C#, ASP.NET Core, Entity Framework, PostgreSQL (Supabase), OpenAI API, LootLocker

