I'm the technical lead of a 17-person engineering team at Blackwired, a threat intelligence company, where I build LLM infrastructure and AI agents. Before that I built machine learning and low-latency trading systems in London, and founded a crypto wallet used by 50,000 people a month.
Does letting an AI agent reason before it acts make it safer? Across 83,584 test runs in two workplace scenarios, the biggest factor was whether the agent had read its rules. Reasoning helped it find those rules and apply them with judgement.
OpenAI's Parameter Golf challenge asks for the best language model that fits in 16 MB and trains in under ten minutes. My entry lets a simple statistical model take over wherever it's almost certain what comes next, which improved the model's score. When a reviewer found an error in my headline number, I published the correction within a day.
Senior Architect · Blackwired (remote)
Technical lead of a 17-person team building LLM infrastructure and AI agents. Cut LLM inference costs by more than $30k a month.
Senior Full Stack Engineer · Kern AI, Bonn
Rebuilt the backend and built LLM routing and document-processing pipelines.
Senior ML Engineer · Siam Investment Management, London
Forecasting models and low-latency trading infrastructure.
Founder · Mimoto.art, London
Built a crypto wallet platform and grew it to 50,000+ monthly users.
Senior Software Engineer · Capital Eurasia, London
Low-latency trading systems for Eurex markets.
Earlier
Zither, Ethereum, Convert Technologies.
BSc (Hons) Computer Science · University of Hertfordshire, 2015
First-Class Honours.
hello (at) anmarhindi.com