Prema Vision · AI products, built and taken to production

If it only works because one person remembers it, it isn't a system yet.

We design and build AI systems for companies and take them to production: inside the iPhone and Apple Watch apps your people use, in the conversations your experts have with clients, and in your daily operations. Your team runs them and changes them without a developer.

Tell us what lives in one head See what we build

15+ years of production software · built first in our own products

Groupon
Autodesk
BigPanda
Routine Run
Structed

Sound familiar?

Six ways people describe the same problem

Different words, one cause: the business got more complex than the system it runs on, and the missing part lives in someone's head.

“We're growing, and I still check every handoff myself.”
Founder, 15 to 150 people
“Three people know how this works. If one of them leaves, half of it stops.”
Expert services firm
“We spent money on AI pilots. The way we work hasn't changed.”
Head of operations or AI
“Our app has users. Siri, the Watch and Apple's on-device AI can't see it.”
Product lead, iOS app
“Every small change goes back to a developer.”
COO
“Our developers built what we asked for, not what we need.”
Founder without a technical co-founder

What we build

AI products and AI systems, one way of building them

Both start the same way: we take the part that has no shape yet and give it a form people and AI can work with. Then we build the product around it and take it to production.

Mobile apps with AI under the hood

Apple now ships a language model inside every recent iPhone. It runs on the device, costs nothing per request and keeps data on the phone. Siri, Shortcuts, Spotlight, widgets and the Action button all read one set of App Intents. Most apps use none of it.

· AI features on the device and on Apple's Private Cloud Compute
· Siri, Shortcuts and the Action button for your most repeated actions
· Apple Watch apps that move the work on from the wrist
· MCP connectors, so Claude and ChatGPT can use your product
· New apps, from zero to the App Store

Built first in Routine Run: 60+ App Intents, Apple's models write a routine from one sentence, and a PagerDuty page starts a runbook on the Watch.

Mobile + AI packages and prices

AI interviews and expert systems

Your best expert can't sit in every conversation, and won't be with you forever. An AI interviewer asks what they would ask, one question at a time, and records the answers as structured data. Your experts decide what it knows and change it themselves.

· Client intake that ends in structured requirements, not a transcript
· Capturing a senior expert's judgement before they leave
· A back office to edit, test, publish and roll back, with no developer
· Quality measured on test runs, not guessed from complaints
· A person approves anything that can't be undone

Already running in Structed, which we can set up for your domain instead of building from scratch: facts clients gave stopped going missing (18 of 19 recovered), and replies went from 20–60 seconds to 1–13.

AI interview packages and prices

AI in your operations

A process mapped, then automated: AI takes the reading, sorting, drafting and routing, connected to your CRM, inbox, helpdesk and Slack. Plus audits and observability for the agents you already run.

Engineering leadership and infrastructure

Architecture and code audits, a technical partner for founders, taking a product back from an agency, and infrastructure that runs without a babysitter.

Proof

Everything we sell, we built first

No slideware. Two products in production, built the way we build yours: one accountable engineer, AI agents doing the routine work, tests from the first day.

Routine RunDenys is the founder

A platform that turns a process people carry in their heads into one that runs. Its native app is the first form; underneath, an API, webhooks and an MCP connector link it to other systems and AI assistants.

60+

App Intents for Siri, Shortcuts and the Action button

Source ↗

2,200+

automated tests

Source ↗

67 ms

to compute the day's schedule on a real phone

was 8.5 s

Source ↗

MCP

connector listed in Claude's directory

Source ↗

Check it yourself

Structedbuilt as CTO

One AI-led interview turns what a client carries in their head into a structured set of requirements, even when they can't quite explain what they need. Experts change the agent and publish without a developer.

18 of 19

dropped facts recovered

Source ↗

1–13 s

per reply

was 20–60 s

Source ↗

¼ fewer

turns per interview, no topic asked twice

Source ↗

1,400+

automated tests

Source ↗

Check it yourself

What people who worked with Denys say

Denys gets the day-one architecture decisions right.

Elena Pinteleeva

Co-founder & Head of Structed Lab · 20+ years in C-level roles

Recommendation on LinkedIn ↗

He is very production-aware, consistently applying best practices such as feature toggles and gradual rollouts to ensure smooth, safe delivery.

Hadar Carmely

Senior Engineering Manager, AI platform & incident management · managed Denys directly

Recommendation on LinkedIn ↗

We originally brought Denys in to strengthen our frontend but he quickly proved himself as a full-stack force, taking on backend, infra and automations with ease.

Juan D. Cardozo

Staff Software Engineer · managed Denys directly

Recommendation on LinkedIn ↗

How both were built

Stack and expertise

What we build with

Every item below is something we have shipped to production, and the line under each group says where.

AI and agents

LangGraph agents and multi-agent pipelines
MCP servers and connectors
Evals: golden sets, test personas, LLM-as-judge checked against people
Tracing and observability in LangSmith
Model routing across OpenAI, Anthropic, Google through OpenRouter

Shipped at Structed, Routine Run

Apple platforms

Swift and SwiftUI
App Intents, Siri, Shortcuts, the Action button
Apple Watch apps and complications
Live Activities and widgets
App Clips

Shipped at Routine Run

Backend

TypeScript and Node.js, NestJS
Python and FastAPI
PostgreSQL
APIs, webhooks, OAuth 2.1
Payments with Stripe

Shipped at Structed, Routine Run, BigPanda

Infrastructure

AWS and Cloudflare Workers
Docker, Kubernetes, Argo CD
Deploys without downtime
CI with tests and AI code review
Sentry, logs and traces, alerts to the team

Shipped at Structed, Routine Run, BigPanda

How a project runs

Five steps, the way a run goes in Routine Run

You always know which step you're on, what it costs and what comes next. Fixed price, agreed before the work starts.

1Free
A 30-minute call
This week
You tell us what lives in one head. We tell you whether we're the right people, and if not, who is.
2Fixed price
The map
1 week
We talk to the people who carry the process and write it down: steps, decisions, owners, what AI can do and what stays human. The map is yours even if you stop here.
3Fixed price
The build
2 to 16 weeks
Priced from the map. A working build every week, on your phone or in your browser, not a slide.
4Handover
Production
Launch day
Live with your real users, with tests, traces, and a short handover doc your team can extend. The code and accounts are yours.
5Included
The check
4 weeks later
We look at what actually happened in real use and fix what the data shows, not what we guessed.

Built for every person who will use it

We design through the eyes of everyone who will use a product, not one imagined persona. A person with ADHD, someone on their first day, an expert in a hurry, someone using VoiceOver. If a design leaves load on the person, we count it as a bug.

One step on screen at a time
Colour never carries meaning alone
Undo instead of “are you sure?”
Plain words over jargon, in the user's language

Who you work with

One accountable lead. A team that grows with the work.

Denys designs every project and reviews everything that ships. AI agents do the routine engineering, and when the scope needs more hands, specialists join under the same review. Victoria makes sure it works for the people who will use it.

Denys Korolkov

Co-owner · lead engineer

15+ years shipping production software, from Groupon's merchant booking platform to checkout for Autodesk's 7M+ users and BigPanda's AI incident tooling. Founder of Routine Run; built Structed's AI interview as CTO. Leads every project and reviews what ships.

LinkedIn denys@premavision.net More

Victoria Von

Co-owner · design and client experience

Product experience, behavioural design and creative direction; tests every build with real people. Co-founder, Sales & Distribution at Routine Run.

LinkedIn victoria@premavision.net More

Notes

What building it taught us

Mobile + AIOctober 17, 2026

Apple put a language model in every new iPhone. Here's what it's good for

What Apple's on-device model did well in Routine Run, where it failed quietly, and the test that keeps a prompt and its parser from drifting apart.

Mobile + AIOctober 17, 2026

Let AI assistants into your product without handing them the keys

How Routine Run's MCP connector lets Claude read runs and act, while every change a model writes waits for a person to approve it on the phone.

AI interviewsOctober 17, 2026

Our AI interview was losing facts clients gave it. 18 of 19 came back

Three fixes from the Structed AI interview: a safety net that matched the wrong thing, a question asked until the turn limit, and a trace lookup that made replies take a minute.

All notes

Why the logo is a labyrinth

Theseus found his way out of the Minotaur's labyrinth because Ariadne gave him a thread.

A process that lives in one head is the labyrinth. Written down step by step, it becomes the thread: anyone can walk it and find the way out, not only the person who built the maze. That is the work.

The thread, on your phone

Five questions we ask on a first call

A routine you open in Routine Run, the app we built for exactly this: a process written down step by step, so anyone can walk it.

A Routine Run routine

First call with Prema Vision

Five things to have in mind before a 30-minute call. Two sentences each is enough.

1
Name the process
The one that only works because someone remembers it.
2
Who carries it today
One person, a few, or nobody for sure.
3
What it costs when it breaks
Hours, money, a client. A rough number is fine.
4
Why now
What changed this month or this quarter.
5
The tools it touches
CRM, inbox, spreadsheets, your app.

QR code that opens “First call with Prema Vision” in Routine Run

Tap or scan with your iPhone

Open in Routine Run

Opens on any phone or computer. To run it, get the app on TestFlight.

Which process in your business lives in one head?

Two sentences are enough to start. Denys reads every message and replies within one business day.

Start a project

If it only works because one person remembers it, it isn't a system yet.

15+ years of production software · built first in our own products

Groupon
Autodesk
BigPanda
Routine Run
Structed

Built for every person who will use it

One step on screen at a time

Colour never carries meaning alone

Undo instead of “are you sure?”

Plain words over jargon, in the user's language