hey, i am damir, ml+bio researcher and cto @ agentmash.
school
nspm almaty · physics and math school
graduated from the best school in the country.
yandex lyceum · programming
the most prestigious programming lyceum in central eurasia.
3%top of the cohort
taekwondo · alongside all of it
master of sport of kazakhstan, captain of the national team.
4xworld championship medalist
6xasian tournament champion
veritas ai · research scholar, data scientist
a stock forecasting tool over s&p 500 history.
build
ratingkz · co-founder
a platform for choosing schools and universities.
2/90my first accelerator
bcc invest · summer swe intern
the bank's first and youngest intern, still at school. ai analytics for fraud prediction, built on what i learned at veritas ai.
3m+bank users
vetra
the first product i built alone.
vetra ai · personal ai assistant
in 2025 i built vetra ai alone. it took the messy messages people send each other and turned them into plans.
it got $200k in infrastructure credits from google, openai, microsoft and cloudflare, grew to 10k+ users, and an investment offer valued it at $8m.
then it hit the wall.
kuramoto
built with a team, as co-founder and cto.
kuramoto · rag over company documents
a rag assistant over a company's documents. every answer cited the exact file, clause and version behind it, and three domain agents covered finance, legal and project management.
94% accuracy across 10,000+ live queries, nps 8.7, under $0.01 per query.
it was acquired by a tech company.
my first exit.
research
after that i reconsidered the values and the ways to have an impact, so most of my time now goes into research. i test whether the metrics a field trusts actually measure what they claim.
ai-augmented solo founders · researcher
107 interviews with founders in 15 countries about where ai covers the work inside a one-person company and where it fails. guided by geoffrey moore and professor poh kam wong.
the paper, pdf ↗34 pagesown-baseline validation · independent
a control that asks whether a published single-cell potency score beats the statistic it is computed from. 39,505 cells, 14 published scores audited, 9 could be placed on the stages and 6 of them beat their own baseline.
own-baseline · the long version
the paper, pdf ↗22 pagescoelacanth · independent
a benchmark for systems that check whether a scientific idea has been done before.
the paper, pdf ↗18 pagestalent
supporting young talents.
oylab · co-founder, advisor
one of the largest youth startup organizations in central asia.
nfactorial kids · jury
a regular mentor at youth startup events, and on the jury of nfactorial kids: pitches from founders aged 9 to 12. crazy.
talentlab · astana hub
helped build it from scratch: a program that invests seven figures in young people. brought in mentors and helped set up the labs, for talent from across the country, especially rural areas.
now
what i am doing these days.
pragma · founder
ai automation infrastructure for 40+ companies on monthly retainers, running on open models on hardware the client controls.
nfactorial · mentor
400+ engineers and founders mentored across four tracks. mentee companies reached 80,000+ users and raised $520,000.
that is the path so far. the next part is being built right now.
everything else is on the work page