oleksiv.com > Sviatoslav Oleksiv > Engineer profile
Sviatoslav Oleksiv
Lead AI product engineer. Production AI SaaS, agents and media pipelines. React, Node, AWS. Krakow, Poland.
Three years as lead engineer of a live AI SaaS, one profitable product of my own, thirteen years in production. I own the whole product: architecture, AI pipelines, billing, infrastructure and the consequences in production.
Engagements
Three shapes of work. A fixed diagnostic comes first, a retainer follows if it makes sense for both sides.
| Lead or fractional engineer for an AI SaaS | Long-term ownership of the product: architecture, AI pipelines, billing, infrastructure, and the plain-language updates for the people who are not engineers. The LIVEDAB engagement below is what this looks like after three years. |
|---|---|
| Production-readiness diagnostic for an AI-built app | Fixed scope, fixed price. I read the codebase, run it, and deliver a written report: what will break first, what is risky, what is fine. Remediation afterwards is a separate decision. Apps built with Cursor, Lovable, Replit or Claude Code are the usual case. |
| AI media pipeline pilot | Image and video generation that runs on a server and survives reloads: Remotion rendering, Replicate and Gemini models, ElevenLabs voices, durable workflows. Scoped as a paid pilot with a working result at the end. |
Write to sviatoslav.oleksiv@gmail.com with what you are building and where it hurts, or book a 30-minute call.
Work
LIVEDAB
| Role | Lead engineer, December 2023 to present |
|---|---|
| Summary | Multi-tenant SaaS that turns event photos into branded AI images and videos, with a CRM underneath. Used by event organizers and brands in Germany. 5,300 commits on the main repo. |
| Hard problem | The AI chain first ran in the visitor's browser, so closing the tab killed the job. I rebuilt it as a durable server workflow. Every AI step retries on its own, every step falls back to its input on failure so the visitor always gets a usable photo, and the result arrives by link and email. |
| Stack | Next.js, React, TypeScript, MongoDB, Remotion on AWS Lambda, Replicate, Gemini, OpenAI, Playwright |
mosfoto.com
| Role | Founder, solo, May 2026 to present |
|---|---|
| Summary | Paid tool that turns any photo into the exact residence-card photo Poland's MOS portal accepts. The preview is free, the finished file costs 19 zł. 211 paying customers in October 2026. |
| Hard problem | The framing has to be identical every time, so the models only find the head and level the eyes. The server computes the crop itself and versions each result. Uploads go straight to blob storage, every write route is bot-protected, and files are deleted after a day. |
| Stack | Next.js, Stripe, Replicate (BiRefNet, BiSeNet), Vercel Blob, PostgreSQL, PostHog |
Personalized sales videos
| Role | Own tooling, 2026 |
|---|---|
| Summary | One command renders a demo video for a single prospect. Playwright records their website, Claude writes a call script, ElevenLabs voices both sides, Remotion renders the MP4. |
| Hard problem | The video must not promise anything the real product doesn't do, so Claude writes against the production voice agent's own system prompt. A script-only mode lets me review the copy before paying for voices and rendering. |
| Stack | Remotion, React, TypeScript, Claude, ElevenLabs v3, Playwright, Zod |
Numbers
| Measure | Value |
|---|---|
| Job success on Upwork | 100% across 18 contracts, Top Rated Plus |
| Hours billed on Upwork | 3,658. Average engagement 180 hours, longest 1,300 and still running |
| Years in production software | 13, since 2013 |
| Commits on LIVEDAB | 5,358 in the main repo |
| mosfoto.com | 211 paying customers in October 2026, solo product |
| Stack Overflow | Answers reached 856,000 people |
| The Pets and Love | A blog I built and run, 100,000 monthly visitors |
Each row links to its source where one is public.
How I work with AI coding agents
Test-first, with a closed loop. Business requirements become tests, the agent writes code against the tests, the tests run, and the loop repeats until green. The tests define correctness, not the agent.
The agent gets rules and structure in the codebase before it writes a line: naming, layering, what must never be imported where. A client’s review of that work is the first quote below.
The same discipline runs in production. In LIVEDAB every AI step is a strategy in a typed registry, so the build fails until each transform has a server implementation, and each step returns its input when the model fails.
Notes
- A durable workflow for a multi-step AI chain. Why a browser-driven pipeline lost jobs, and how a server workflow with retries fixed it.
- Every AI step falls back to its input. The one rule that keeps a generation pipeline useful when a model fails.
- Rules for coding agents in a production codebase. Tests first, rules in the repo, and checks that fail the build instead of a review.
What clients said
“I've worked with a lot of freelancers on Upwork, and most of those contracts end pretty quickly. Sviatoslav is one of the few I kept on for close to a year, which tells you most of what you need to know. He owned the front end and API of our public data platform. […] He also put real thought into working with AI coding tools, setting up rules and structure in the codebase so the generated code actually fit.”
Emiel Cockx, public data platform, 650 hours, 2025 to 2026
“One of the best freelancers I've worked with on Upwork. It's a pleasure working with Sviat - he is highly skilled, reliable, always responds quickly, and brings a very positive and friendly attitude to every project. Communication is excellent, and you can always count on him to deliver high-quality work.”
Hannes Meine, web application code review, 2025 to 2026
“Sviat is great to work with. He comes with ideas on how to help improve the project. He works diligently. I've worked with him on 2 separate projects now. He continually seeks to improve his skills and deliver above expectations on any project.”
Michael Ochs, fintech platform on Azure, 307 hours, 2024 to 2025
“Sviat was a pleasure to work with - professional, reliable, and incredibly fast without compromising quality. I'd gladly hire him again.”
Adrian Hanrath, LIVEDAB, CRM MVP on React and NestJS, 2025
Verbatim from Upwork reviews; a cut is marked with an ellipsis.
Contact
| sviatoslav.oleksiv@gmail.com | |
| Call | Book a 30-minute call |
| linkedin.com/in/oleksiv | |
| GitHub | github.com/oleksiv |
| Upwork | upwork.com/freelancers/oleksiv |
| Location | Krakow, Poland |
| Languages | English, Ukrainian, Polish |