Senior Software Engineer at Google DeepMind (Gemini). Based in Zürich.
I build systems across the entire stack — from GPGPU kernels to mobile apps — and I spend my free time training language models from scratch to build intuition about where AI is heading.
🔬 LLM Architecture Research — Training and evaluating transformer architecture improvements at 150M-parameter -> 1B scale. Exploring training efficiency, novel attention mechanisms, and hybrid architectures.
🤖 AI Agent Systems — Building autonomous coding agents and multi-agent orchestration frameworks.
| Project | Description |
|---|---|
| protokit | High-performance Protocol Buffers for Rust. Only implementation supporting full TextFormat spec. |
| termineer | Autonomous AI coding agent — task decomposition, multi-agent orchestration, universal tool abstraction across Claude/Gemini/GPT. |
| genagent | Multi-provider AI desktop assistant. Rust backend + Swift frontend, cross-platform FFI. |
| stor | Typed embedded key-value database on RocksDB with zero-copy deserialization. |
| mml | Micro ML library — automatic differentiation engine targeting Apple Silicon CPU + GPU. |
- Training efficiency — the transformer paradigm has a decade of scaling left, efficiency is the bottleneck
- Compression & intelligence — the Hutter prize framing, what it means for model design
- Game-theoretic AI — poker AI, reasoning under uncertainty
- Interpretability — understanding what's happening inside models, not just what comes out
Rust · Python · Go · C/C++ · Swift · TypeScript · Kotlin · Java
PyTorch · MLX · CUDA · OpenCL · Kubernetes · gRPC/Protobuf