Skip to content
Prema Vision
ServicesMobile + AIAI interviewsWorkNotesAbout
Start a project
Prema Vision

AI systems for companies, designed, built and taken to production: in your apps, your client conversations and your operations. Retainers keep a person on it after launch.

denys@premavision.net

ServicesMobile + AIAI interviewsWorkNotesAboutRetainersHow we work togetherContact
For local businesses around Palm CoastRoutine RunDenys Korolkov, korolkov.proDenys on LinkedInVictoria on LinkedInPrema Vision on LinkedInPrema Vision on Facebook

© 2026 Prema Vision LLC · Palm Coast, Florida · Remote across the US · Privacy

Work

Two products, milestone by milestone

Prema Vision shows the products its lead engineer built and runs, because that is where you can check the work: open the app, read the changelog, ask how a number was measured. Newest first, like a changelog: every milestone has a date, and most have a link.

Routine Run · since August 2026 · Denys is the founder

A process people carry in their heads, turned into one that runs

In Routine Run a process is described once in a sentence, walked through step by step, and improved run by run. The native app is its first form; underneath is a platform that connects processes to the systems and AI assistants a business already uses. No account, private by default.

  1. NextNext

    The same routine in any browser, then iPad and Mac, then beyond Apple.

  2. With the releaseNext

    App Clip pilots with local businesses: a code at the door, the steps on the visitor's phone.

  3. In progressNow

    Getting version 1.0 onto the App Store, for everyone.

  4. 14 October 2026

    Connections: any system or PagerDuty can start a run through its own address. Claude and ChatGPT connect with a code, and what they write waits for the person's yes. The MCP connector is listed in Claude's directory. Source ↗

  5. 4 to 5 October 2026

    A routine scanned at a door opens in an App Clip with nothing to install; a routine can be published under a code, a poster or an NFC tag. Source ↗

  6. 2 October 2026

    Assistive Access: one large step with Next and Not now. Larger Text and Reduce Motion reach every screen. Source ↗

  7. 29 September 2026

    Apple's Private Cloud Compute writes routines inside the app on iOS 27. The person chooses: Apple's cloud, or only this phone.

  8. 24 September 2026

    Apple's on-device model writes a routine from one sentence. Siri can change a routine, and one Undo puts it back. Source ↗

  9. 22 September 2026

    A real PagerDuty page started an on-call runbook on the Lock Screen and the Watch, about two seconds from event to decision. Source ↗

  10. 19 September 2026

    Private iCloud sync, earlier versions of every routine, and support back to iOS 26.1 and watchOS 26.

  11. 16 September 2026

    Public beta builds on TestFlight, with every build described in plain words. Source ↗

  12. August 2026

    Routine Run founded. The method Denys carried for years becomes a native Apple app: the engine and App Intents first.

60+
App Intents
Source ↗
2,200+
automated tests
Source ↗
67 ms
day's schedule on a real phone
was 8.5 s
Source ↗
~2 s
from a PagerDuty event to Routine Run's decision
Source ↗

Check it yourself

  • routinerun.com ↗
  • TestFlight beta ↗
  • Every build, newest first ↗
  • The method ↗
  • MCP connector ↗
  • Claude directory ↗
  • API and webhooks ↗

Structed · December 2025 to October 2026 · built as CTO

One AI interview instead of weeks of back-and-forth

Structed turns one AI-led client interview into a clear, structured set of requirements, even when the client can't quite explain what they need. Experts run it themselves: they edit the domain taxonomy in a sandbox, test it and publish to production with no developer. Engineering ran as one engineer working with AI agents: Claude Code and Copilot on the repository's instructions and 28 skills, with AI code review in CI.

  1. 17 September 2026

    Demo day in London, with the product able to show its own measurements. Source ↗

  2. August 2026

    Self-serve sign-up with a trial, organizations with seats, and test personas run against the live stack and compared run against run. Selected for the Grishin Robotics accelerator, cohort 4. Source ↗

  3. July 2026

    Paid accounts through Stripe, a moderation step before requirements reach the client, deploys without downtime, and a security pass.

  4. June 2026

    The domain taxonomy moved out of code into a database with its own admin, and an MCP server opened interviews to AI assistants.

  5. April 2026

    Simulated clients: an AI persona runs an interview end to end, so a change is tested before a real client sees it.

  6. March 2026

    Traces on every run, model routing, and prompts domain experts can edit without a developer.

  7. February 2026

    Audited the base, ended the outsourcing and took over all engineering.

  8. December 2025

    Designed it with the founder from a blank page: the architecture and the agent concept. An outsourced team started on the base.

18 of 19
dropped facts recovered
Source ↗
1–13 s
per reply
was 20–60 s
Source ↗
¼ fewer
turns per interview
Source ↗
1,400+
automated tests
Source ↗

Check it yourself

  • Try structed.ai ↗
  • Milestones in full ↗
  • Four weeks to a demo day ↗
  • Grishin Robotics accelerator ↗
  • Denys on LinkedIn ↗

Before AI · 15+ years

Three systems that carried real load

What you get from them is judgement about where things break.

Groupon

Main frontend contributor on the merchant booking platform while its team grew from 2 to 8 engineers.

A product merchants depend on every day has to stay easy for the next engineer to change.

Autodesk

Checkout for a product with 7M+ users, a system under real load.

Respect for the invisible failure: in billing it shows up on an invoice, weeks later.

BigPanda

AI incident tooling for enterprise on-call teams.

Noisy input quietly breaks every decision below it, so the signal gets fixed first.

Start a project