I am a software engineer and trader. I build systematic trading systems end to end, from the ML that looks for an edge, to the engine that trades it live, to all the infrastructure in between. Markets pulled me in because they give you the hardest kind of feedback. You are either right or you are not, and the P&L tells you which.

I mostly work in Python. I build trading engines, agent systems, and data pipelines. If something is worth doing twice, I would rather build the thing that does it a thousand times on its own.

I also build cloud infrastructure when the workload calls for it. One example is a scalable genomics processing environment on AWS. I used Karpenter to autoscale the compute to the size of each job, and Lambda for the event-driven glue that holds the pipeline together.

KuroAlpha is my own trading platform on OANDA, and it also serves as my ML workbench. I started it in 2021, and I have been running my own capital through it since 2023. It is where I train models, test ideas, and risk my own money to find out which ones actually hold up. EURUSD, US30, and GER30 are my main markets. I build a model, deploy it, break it, refine it, and repeat.

I co-founded a stealth startup and built the agentic harness behind it. It is a multi-agent system staffed like a real quant desk, with analyst, trader, and developer roles.

The agent work is where I lose track of time, and it goes well past trading.

AERARIUM is an agent that plays games straight off the screen, without reading game memory. It looks at the screen, builds a picture of the world it has been dropped into, plans, acts, and re-grounds itself when the game changes underneath it. I built it to watch how an LLM handles a problem it has never seen before. Where it guesses, where it falls over, and how it recovers. Pixels in, decisions out.

ARENA is a world I built for agents to learn in. They try things, keep what works, build up a skill library, get critiqued, and go again. If AERARIUM is the eyes, ARENA is the learning.

Most of this lives in private repos, but I am happy to walk through the architecture. Just ask.

I am comfortable across the modern LLM stack. I use smolagents and LangChain/LangGraph for orchestration, RAG when retrieval matters, and fine-tuning when prompting is not enough. I rely on evals so I know whether a change actually helped. I use vLLM and Ollama when I want to run models close to the metal, and I build from primitives when nothing off the shelf fits, which on the harder problems is most of the time.

I have wanted to run my own fund for as long as I can remember, and it is slowly coming together. Further out, I want to fund research teams of my own, working on problems where the engineering and the science both have to be right. Finance and genomics are the two I have fallen for so far.

Quant finance, ML, and systems design are the foundations. Right now I am focused on multi-agent coordination, evals and tool-use, ML for drug discovery and genomics, post-training of frontier models, and high-throughput inference on modern accelerators. There is always something new I want to understand next.