Beiimbet Sarsekeyev

I build production ML systems that handle what most engineers never encounter. Sensitive, air-gapped environments where data can't leave the building and infrastructure must be built from scratch. For the past five years at Tackle AI, I have taken document AI from a research idea to a running production system, owning the full lifecycle: I frame the problem, design and train the models, build the infrastructure to serve them, and keep them running in places with no internet and no margin for error. Those years shaped how I solve problems. With no cloud to fall back on and sensitive data closing off the easy paths, I learned to work from first principles: when the standard tooling did not fit, I built my own, and when the obvious fix stalled, I dug until I found the real bottleneck. That is the instinct I bring to a hard problem now. The through-line in my work is judgment about where AI breaks. In high-stakes domains, the answer that quietly reaches a client is the one that matters, so I build systems that know when to doubt themselves and ask a human. I have turned months of manual document review into work that finishes in seconds, and I trust a model in production only once evaluation proves it is actually better, not just newer. I care about building things that hold up in the real world, on messy scans and edge cases that clean benchmarks never show. What I bring is a way through complex, ambiguous problems, backed by years of solving exactly those in a startup, end-to-end and under real constraints. If that is the kind of challenge you are facing, whether it is trustworthy AI, sensitive data that cannot leave the building, or owning the full pipeline end to end from idea to production, I would like to hear about it.