MyLexilingo
A language-learning platform. Teachers and learners bring their own vocabulary and get exercises, conversations and stories at the right level. My main independent project.
Independent projects
Some of these are real products, some are puzzles made for fun, and many are experiments with AI. They're grouped below.
A language-learning platform. Teachers and learners bring their own vocabulary and get exercises, conversations and stories at the right level. My main independent project.
Describe an idea you half remember from a podcast, read the matching passages in context and jump to the exact moment in the recording.
Follows asking prices of Ford Mustangs across Dutch marketplaces, with price history for each car.
A daily Macedonian word puzzle: swap letters to complete intersecting five-letter words in 15 moves.
A Macedonian Spelling Bee: make words from seven letters, always using the centre one.
A playable tribute to a racing game from my childhood, built with a coding model from two prompts. The prompts are published.
An open language model with a tool-calling loop, running entirely in your browser through WebGPU.
Generations from open-weight models, published together with the prompts so you can see how they were made.
Build Solara components from a single HTML file connected to Python state.
A SQLite workbench in the browser, with optional Gemini-assisted queries. The database stays on your machine.
A small Python framework for AI agents, designed so the instructions each model receives are easy to inspect.
Demos that make token probabilities visible and show how to use them for classification and retrieval.
A Python library for sending large numbers of asynchronous OpenAI requests from scripts and notebooks.
Dictate code and ask questions inside JupyterLab. Built together with Vitalii Karniushin.
More small repositories and experiments are on GitHub.