Intro
Azamat Erkinov
I build AI systems for data that can't move, and I think most companies are renting infrastructure they should own.
I study data science and economics at Minerva University, and I build AI systems for data that can't move — either because it's private, or because it's public and nobody's made it usable.
Right now I'm a forward deployed engineer at Tensor AI, putting quantized open-weight models on enterprise hardware so companies stop sending their documents to someone else's API. Before that I automated FCC compliance testing at UL Solutions. Since early 2025 I've been building an assistant for Safecast — it lets anyone ask questions in plain language across 200 million radiation measurements collected by volunteers since Fukushima.
I won “Most Impressive Engineering Lift” at a YC-RFS hackathon in Tokyo1 for a multi-agent research platform, and I ran an essay contest on AI safety that reached 900 people across 40 countries.2
I'm Uzbek, I work in three languages, and Minerva has me living in a different city every semester. Currently in San Francisco.
Say hello: azamat.erkinov@tensorai.io, or find me on GitHub and LinkedIn.
Natural language over 200M radiation measurements
PRODUCTION · 2025–present
An MCP-based assistant embedded in the Safecast map, answering plain-language questions across three data services.
Cutting inference cost 62% without data leaving the building
DEPLOYED · 2026
Quantized open-weight models on-prem, client-specific RAG over 400K+ documents, and governed MCP tools.
10-K research loops cut to 12 seconds
WON · 2026
Multi-agent hedge-fund research platform; won “Most Impressive Engineering Lift.”
Writing