Oleg Malkov

Hands-on AI Product Builder | ex-Property Finder | Rapid LLM prototyping

After years leading engineering at scale, I'm choosing to go deep with one team, building and shipping fast.

Based in Dubai. Solo, self-directed work since August 2026. Each project states its stage.

Track record

Property Finder, Dubai Oct 2020 – Jul 2026

Introduced BDD living specifications: every feature starts as user scenarios, and they became the company's primary record of what the product must do.
Ran delivery for my teams in Scrum and Kanban, so product and engineering managers could focus on product, tech and stakeholders.
Billing for about 20,000 real-estate companies (about $200M a year): designed the ledger and ended duplicate charges as a cause of billing incidents.
One platform for customers instead of three: cross-checked the data across eight sources and exposed the corrupted records that blocked the migration.
Frontend foundations: Next.js for the consumer site, and React Native for the B2B web and the native iOS and Android apps.
Led coding-agent adoption (nearly 200 licences) and AI code review that cut expert review effort by about 80%.

ONE.COM, Dubai Nov 2015 – Oct 2020

Built the website builder rolled out to about 1M customers, including the paid webshop product inside it.

How I build with AI

Human defines trajectory, agents do everything else. I design, direct coding agents, and review and test what ships. With AI the speed of iteration is what matters. Failure is expected, so every loop is built to teach me something fast.

I start from the customer outcome and work backwards, as in Amazon's Working Backwards: what the user gets, the number that shows it worked and the result that would prove me wrong, on one page, before anything is built.
Agents are coworkers on a Kanban board: work flows in small pieces, because AI iterations are too short for sprints.
Every brief says what done looks like before work starts, written as user scenarios.
Two different AI models design the same thing without seeing each other, then argue it out, and the evidence decides.
The one who builds is never the one who checks, and I trust a check only after I have seen it catch a deliberate mistake.

Building now

in development, testing with real users

From styling to wardrobe digitisation, for my wife's business

Find the unmet need, then build it. After 14 years as a mother of three, my wife is making her dream of working independently real. Her business grew from simple styling into wardrobe digitisation, and then into an app that manages both the client's wardrobe and the stylist's work.

The software was the easy part. The photos were hard: 13 stand and hanger designs in five weeks, and fixed daylight lamps because no AI fixes a colour the camera never captured. What worked: an invisible half-bust hanger that gives the garment a female shape, flat-backed to slip in and out. Hangers sold online show in photos and the one invisible model was flat, so we printed our own.

The photo station →

Also built

in use

A home workshop for my kids, built for what comes next

AI is changing knowledge work fast, so the kids learn by making. Making: 3D printers, laser cutters, a vinyl cutter with hundreds of heat-transfer vinyls, an embroidery machine, a heat press and a full set of hand tools. Materials: concrete, gypsum, two-part resin, candle wax and about 100 kinds of stone beads for bracelets. Programming: a spider robot for my 14-year-old and a robot dog for the 8- and 5-year-olds. Some of it runs only with an adult beside them. We hold standups with the kids: how they feel today, what they made, what's next, where they're stuck and how we can help.

in use

Private messaging app

Phone-first chat threads with live updates and notifications, on my own hosting, used from my phone every day.