Vladimir Kisselev
Software engineer in Toronto.
I have been building software since 2018, usually as the only engineer on the product, sometimes as the person who put the team together.
Most recently TruForagers: a 29-module Kotlin Multiplatform stack spanning Android, a React web client and a Ktor server, sharing one domain model across all three targets, built end to end in four months with 1,400+ tests.
Before that I formed and mentored a five-engineer team across Android, iOS and backend, and built the Android app it shipped from an empty project to the Play Store. Earlier: contract front-end work in Canada and South Korea, a WordPress storefront turning over about $1M a month, and an iOS MVP I co-founded and shipped to the App Store.
Lately I spend most of my attention on a narrower question: how do you get an agent to produce work you would actually ship? The short answer, and the thing I got wrong for months, is that the checks have to live outside the model.
Writing
-
The gate goes outside the model
Design notes from running an engineering agent unattended for six weeks. Every lesson pointed the same direction: wherever I trusted the model to hold state, follow a process, or check its own work, that responsibility had to move out of the model and into something deterministic.
2026 -
LLM slop is a skill issue
The context window is not a bucket, it is a focus override. Your prompt is a compression of an infinite idea, and the gap is where the model guesses. The fix for both is a process with a fresh context at every stage.
2026
Open source
-
claude-asimov · MIT
The extracted agent runtime. The verification gate runs outside the model, so it can't be skipped. Over roughly six weeks of unattended operation the loop authored 1,127 of the 1,243 commits in the codebase it drives, about 26 a day, every one past a lint and test gate before landing. Source.
-
agents-skills
Five Claude Code agents and twelve skills implementing a research-to-execution pipeline: recursive fact-finding, question generation through four analytical lenses in parallel, research triangulated against what the code actually does, then design, technical structure, and a vertical execution plan. Source.
Experience
-
TruNorth Naturals · 2026 to now
Sole engineer on TruForagers, a traceable wild-mushroom supply chain: Kotlin Multiplatform across Android, React web and a Ktor server, with a QGIS-backed geospatial ingestion pipeline underneath it. Bilingual from v1.
-
Private practice, contract · 2021 to 2024, 2026 to now
NareTrends SmartFarm in South Korea: rebuilt a legacy IoT agriculture interface as a responsive browser and tablet app on Vue and PostgreSQL, localized for market entry, and presented it to the Ministry of Agriculture to win a trial deployment at an agricultural university. Since then, marketing sites for Artlok Logistics and two Toronto businesses on React 19 and TypeScript.
-
Production Reserves · 2024 to 2026
Built a youth-sports platform's Android app from scratch to production, on multi-module Clean Architecture with real-time chat and maps-based gameplay, reaching 2,000+ installs and ~99% crash-free. Formed and mentored the five-engineer team around it.
-
E-commerce retailer · 2019 to 2020
Ran a WordPress storefront doing roughly $1M a month in online sales, and the IT infrastructure under it.
-
Happy Hour Labs · 2018 to 2019
Co-founded the company and was its only iOS engineer: chose the stack, defined the architecture, and shipped the MVP to the App Store.