I build mobile apps, developer tools and the low-level pieces under them: networking, on-device ML and Rust. Some of it is open source, some under Dockito Labs.
- An AI-first docs engine: a place where docs, sheets, slides and API specs are written and kept up to date by people and AI agents together.
- Ternary LLMs: researching whether models with ternary weights can beat 16-bit ones at the same memory, starting with small GPTs.
- PocketRocket: Claude Code running on an Android phone. A fork of Termux with a chat screen, several sessions at once and separate git accounts per workspace. Early and experimental, under Dockito Labs.
- eywa: a local knowledge base with hybrid search and reranking, in one Rust binary
- transformers: multi-agent orchestration for Claude Code
- adx: connect to Android devices over WiFi and mirror them with scrcpy
- flutter_quic: QUIC for Flutter, using the Quinn Rust library (pub.dev)
- inference: ML inference in Flutter with Rust engines, Candle and Linfa (pub.dev)
- quicrtc: real-time communication in Rust over QUIC and Media over QUIC
Networking (QUIC, WebRTC), Rust, running ML locally, and getting phones to do more than they usually do.
Mostly Kotlin, Dart and Rust, with some Swift and TypeScript.




