I build AI systems and trading systems.
I run Zobyt, a finance-and-AI software consultancy, and I have been writing open source and teaching for most of my career. Most of my work sits where careful engineering matters more than demos: LLM and agent systems, machine learning, and the execution and data infrastructure behind automated trading.
Two domains I work in, sometimes together
AI systems
Applications and infrastructure built on language models and machine learning, for any domain.
- LLM applications, agents and tool use
- Retrieval (RAG), evaluation pipelines and guardrails
- Workflow engines that compose several kinds of models
- Machine learning and reinforcement learning, PyTorch
- Getting prototypes to production, and more
Trading and fintech systems
The software that lets a strategy run unattended, built inside the client's own repository.
- Execution engines and order management
- Exchange and broker integration over WebSocket and REST
- Backtesting and strategy-evaluation infrastructure
- Market making and on-chain execution
- Risk controls, monitoring and kill switches, and more
Selected work
All work- MXGo.aiOpen-source AI agents for email. Forward a message; an agent plans, researches with tools and replies.AI systems, open source
- Execution engineOrder management for US equities that survives gap opens, crashes and reconnects, behind a kill switch.Trading systems
- LLM workflow engineComposed several model types, including LLMs, into production pipelines at an AI-infrastructure startup.AI systems
- Backtesting suiteReplays history with explicit fill, slippage and latency assumptions, and measures the backtest-to-live gap.Trading systems
- wtfpythonExploring Python through surprising snippets. Endorsed by Andrej Karpathy, used in university courses.Open source
Have something worth building?
I take on a small number of engagements at a time, directly or through Zobyt when a project needs a team. Tell me what you are working on.