San Francisco
I founded LFCarry and ran it for about a decade. For the last three years my actual work has been narrower and stranger than “founder”: I build fleets of autonomous AI agents and run them in production against jobs that used to need people.
I started the company in 2016 as LegionFarm — a marketplace where people hire skilled players to coach them or play alongside them — and it now operates as LFCarry. It went through Y Combinator in the Winter 2020 batch, and TechCrunch covered the rounds along the way. I have since stepped down as CEO and stay on the board.
What I do now is not demos. The systems talk to customers, price things, catch their own mistakes, and write their own post-mortems when they get something wrong. The interesting part was never getting a model to answer. It was building something that can tell the difference between what is established, what is inferred, and what someone wants to be true — because at volume, a system that cannot make that distinction produces confident nonsense faster than anyone can check it.
A marketplace for hiring skilled players as coaches and teammates, founded as LegionFarm. Y Combinator W20. The company profile reports cumulative scale on the order of $45M in sales and 250k+ customers — lifetime totals it supplies itself, not a current run-rate.
A governed multi-agent system operating against the live marketplace: buyer conversations, pricing, catalogue integrity, payouts, outbound. Two of the services are documented and running in production; the wider fleet is a designed architecture, and I keep that distinction in writing rather than rounding it up.
Each agent maintains its own written record of what it learned and what it got wrong. That log is the actual product — a fleet whose mistakes are not visible cannot be improved.
Built to brief hackathon organisers on who was walking through the door. It enriched 13,733 profiles, each pass spending forty or more tool calls before it wrote a word. Aggregates only — the per-person records are not published anywhere.
A repeatable instrument rather than a one-off investigation: parallel subagents across several models, web retrieval, and an independent second rail used to contradict the first. The output is a decision plus the one cheap measurement that would prove it wrong.
A research instrument built over the complete back catalogue of a documentary show: every claim extracted, classed as fact, inference or speculation, scored against an auditable rubric, and checked against primary sources. Citations deep-link to the exact second of source video, and the findings that failed are written down next to the ones that survived.
Client and partner work — product, scoring and research systems built for other companies — is not listed here. Those engagements have no public write-up I can point at, and a credential I cannot hand you a document for is not a credential.
I trained as a geophysicist before any of this, and astrophysics and quantum physics stayed with me as a hobby rather than a job. Alongside that I have a long personal practice — meditation, and a serious interest in consciousness. I do not treat the two as separate.
My conviction, stated plainly as a belief and not as a finding, is that the next frontiers are not ahead of us on a straight line — they are where other intelligences may already have been. If genuine information transfer is real, the jump is not incremental; it is the whole curve. That is why I care about the subject rather than merely finding it interesting, and it is why the research above is built the way it is. A field this contested does not need more enthusiasm. It needs someone willing to write down which of their own claims did not survive.
The paragraph above is motivation, not evidence. It is here because you should know why I chose the subject — not as something to take on my word.
The last line is here on purpose. My name has been transliterated several ways over the years, and a person is easier to resolve when the variants are stated rather than inferred. Separately: beliankin.com is not me. That is my brother Kirill, who runs his own studio. We are different people and the records should not be merged.