Hi! 👋 I'm Iskren.

I untangle expensive technical messes.

Legacy systems nobody wants to touch. Migrations that have been "next quarter" for two years. AI features that sound great in meetings but never ship. I go into codebases other people avoid and get the team shipping again. Solo when one engineer is enough. Leading a small team when it isn't. Either way, one person is accountable for the outcome.

What I do

🔧 Rescue & migrate

Systems everyone is afraid to touch, made boring again.

Legacy-to-modern migrations: 5 companies migrated, none of them stopped shipping.
Docker/Kubernetes clusters - reliable, documented, maintainable by your team after I leave.
CI/CD pipelines that ship code safely and predictably.
Cloud costs sized for actual usage, not theoretical peaks. Starts with an audit.

Case: Salesforce-embedded mobile app - two years of migration, zero feature freeze (2.5 million users)

Cordova to React Native, Angular.js to React. Rather than a freeze-and-rewrite, I designed a plugin abstraction layer and a way to render Angular.js inside React, so the team kept shipping features while we rewrote the core underneath them. I also wrote a MITM proxy that swapped the app's JS without a deploy, cutting the feedback loop from 20 minutes to seconds.

⚡ Build

One person who owns the whole picture, from database to deployment.

Full-stack web apps, shipped and running, not handed over as a prototype.
React Native apps - one codebase, both platforms, one bill.
API integrations that work as documented and stay working.
Monitoring that catches problems before your customers do.

Case: Social network for pharmacists - full rebuild in 8 months, zero-downtime cutover (~40K users)

Built from the ground up while leading a team of 4, including migrating and syncing data from the legacy platform, then cutting over live users without an outage window.

🤖 AI automation

Most AI features die between demo and production because nobody decided what happens when the model is wrong.

Evals before deployment, so "it works" is a number, not a vibe.
A human-review path for low-confidence output, instead of silent wrong answers.
Cost and latency measured per document, not per demo.
Data pipelines built to fail gracefully instead of breaking at 3 AM.

Case: Client onboarding - from days to hours, shipped in 1 month (~500 K documents)

Built an AI document-parsing microservice with Claude and Textract. 82% of documents cleared automatically; the rest routed to a review queue with the extracted fields pre-filled, so even the failures were faster than the old process.

Tech

Daily drivers: TypeScript, React, Node.js, React Native, Kubernetes, AWS. Python for data and AI work. Kotlin, Swift, Go or Rust when the job needs them.

How I work

I write code like I'm going to debug it half-asleep at 2 AM - clear, tested, documented.
Today's clever hack is tomorrow's incident.

I take long-term contract engagements, usually 6 months or longer.
Remote from Bulgaria (CET/EET, overlaps with EU and US East Coast mornings).

Got a technical mess that needs untangling?
Let's talk. I'll tell you straight whether I can help, how long it'll take, and what it'll cost.
Booked until the end of 2026. Taking inquiries for Q1 2027.