hi, I’m
also written Дмитрий СеменкевичДзмітрый Семенкевіч
Fifteen years building software where a wrong number costs real money: payments, a device trade-in platform in 16,000 stores. Now Adlega, an AI CFO with one rule: the AI never does the math.
Between releases I poke at the digital world: experiments, new tools, pop science rabbit holes. This site is my workshop.
I am from Belarus, so even the figurine is carved the way our old dolls are. The ornament behind it is a different Belarusian pattern every single day. Thank you for stopping by the workshop. Come along, the digital world still has plenty left to find.
A kenar. An absurdly good singer. He is the reason I get to catch my breath between whatever Zoom has already planned.
My Weimaraner and the most relentlessly energetic creature I know. He loves people, he loves hunting and he trains my nerves daily. He turns up frogs, hedgehogs, hares and brand new adventures faster than I can ship. That is, more or less, the pace he expects from me.
it comes alive now and then. the marks tell you who is who.
I took a random sample of 30000 npm names and rebuilt the dependency graph for every year from 2011 to October 2026, with a GIF of it growing. Packages went from about 2800 to 4.45 million and edges to 19.4 million. The jump of 2024 came from packages created that year, many of them spam that depends on spam. The 100 most used packages get half the share of edges they had in 2015, but the top account still reaches 25 percent of all packages and 14 to 19 accounts reach half.
October 31, 2026 Benford's law flags 84% of the 2025 annual reports. A perfectly honest source would get 60%I ran Nigrini's first digit screen over 5399 annual reports filed with the SEC in 2025. It calls 83.9% of them nonconforming. An honest Benford source of the same sizes would get 60.3%. The flag did not point at later error corrections. To catch 60% of reports 70% of their numbers had to be invented.
October 26, 2026 My control read the file 4 of 4 in August. In October it said the file did not exist and it was rightIn August models with the tools taken away never said the tools were missing. A control with tools read the file 4 of 4. I added the other side: the same questions about a path that does not exist. On Claude Code 2.1.292 the control failed first, because Read was reading the disk of another host. In the clean run both models were right 10 of 10 and haiku handed over the value of a neighbouring file 10 of 10. With the tools off the replies were the same whether the path existed or not.
an AI CFO for SaaS founders
Describe your business in a chat and get a working financial model: revenue, burn, runway, what if scenarios. Minutes instead of weeks in a spreadsheet.
The design decision that defines the product: the AI never does the math. Agents interview the founder, fill the model, explain any number and run goal seek (“how do I triple my MRR?”), but every figure comes from a deterministic engine, tested to stay within 0.5% of the hand built Excel model it replaced.
not one of these numbers was computed by AI: a deterministic engine, like in Adlega.
Terminal tool that tells look alike processes apart and names the ports they hold.
alive90+ browser utilities; no servers, no accounts, files never leave your machine.
aliveCo-founded; a mobile app that ran for a decade. It never became a business, the lessons stayed.
dead