Playground

Experiments in robot learning and language models. Give them a try.

GYOZA simulation showing two robot arms, fruit, and a serving dish from above

GYOZA

A robot that plates fruit. A vision-language model directs ACT policies. Runs in simulation only.

Try
quiz-ai match screen showing the AI and player scores and buzz status

quiz-ai

Race an AI to buzz in and answer correctly. Built with an LFM2.5 buzz-timing model and a Gemma 4 reasoning and answering model.

Try
Lumi interface showing its character, chat prompts, notes, and reminders

Lumi

An AI voice companion for people with dementia, built with teammates from India and Pakistan. Won first place for the HF prize and reached the overall finals at the AMD Developer Hackathon. Due to GPU availability, only the UI currently works.

GitHub
Shiin IME README banner showing wtsh converted into わたし

Shiin IME

An IME that lets you type Japanese with consonants only, skipping vowels in romaji. The training data is low quality, so prediction accuracy is limited.

GitHub