01 / In development
kindoki.app
Founder · Current work
A shared social world for AI personalities.
A social platform built around AI personalities. People can interact with them, follow the thoughts they share, and see how they respond to activity on the site, the news and events in the wider world.
The ambition is a shared social world where each personality carries its own identity and memory across interactions with people and other AIs. Building that experience means deciding what they remember, when they participate and how their behavior develops over time.
As founder, I lead the product’s direction and development, bringing the social experience and the systems behind it together.
02 / Development paused
NeuralRanch
Independent project
Exploring reinforcement learning for trading strategies.
NeuralRanch explores how to make reinforcement learning accessible to people investigating trading strategies. The idea is to let someone configure an experiment, train an agent through simulated market decisions, and evaluate how it behaves on unfamiliar data and in paper trading.
The interesting problem goes beyond producing an impressive backtest. A strategy can fit yesterday’s market and fall apart tomorrow. Trading costs, changing conditions and risk limits all affect whether an experiment tells us anything useful.
The work spans the product concept and the tools supporting those experiments: training, historical evaluation, saved models, execution-cost modeling and paper trading. The ambition is to help people investigate ideas and understand why an agent behaves as it does.
03 / Coauthored research
ChessLLM
UQO
Coauthored research comparing language models on chess puzzles, with a public dashboard.
How well do language models solve chess problems? We compared five models across 1,400 puzzles and built a public dashboard to explore the results. My contribution included the evaluation pipeline and the application presenting the findings.