--- title: CV — AI Architect Final Source project: revenue-strategy type: output status: ready generated: 2026-06-11 --- # Igor Tarasenko **AI Architect / Agentic Engineering Lead** · Amsterdam, NL Email: igor@tarasenko.dev · Website: https://tarasenko.dev · GitHub: https://github.com/Saik0s · X: https://x.com/saik0s ## Summary AI architect and senior software engineer with 16 years building production software across mobile, platform, payments, developer tooling, and applied AI. Ex-Uber. Currently lead iOS engineer and internal AI enablement lead at DNA.inc, helping engineering teams adopt agentic development workflows that hold up against real codebases. I work at the intersection of iOS, LLM systems, agentic tooling, and production reliability: evals, guardrails, MCP tools, on-device inference, AI coding workflows, and pragmatic architecture that turns prototypes into systems teams can maintain. ## Architecture strengths - **Production AI systems:** LLM integration, structured outputs, eval loops, guardrails, cost/latency trade-offs, fallback design. - **Agentic engineering:** Claude Code/Cursor workflows, MCP tooling, agent-readable repo conventions, build/test/preview feedback loops. - **Mobile AI:** Swift, SwiftUI, TCA, UIKit, on-device Whisper, App Store constraints, privacy-first architecture. - **Developer experience:** internal frameworks, automation, build systems, codebase legibility, onboarding, team enablement. - **Applied generative media:** ComfyUI pipelines, Stable Diffusion / diffusion-video workflows, local model operations. - **Scale judgment:** Uber platform/payments background; systems where reliability, cross-team ownership, and failure modes matter. ## Experience ### DNA.inc — Lead iOS Engineer & Internal AI Enablement Lead iOS engineering while also acting as the practical AI adoption lead for engineering workflows. - Lead work on production iOS systems in a large enterprise SaaS environment. - Run internal AI enablement sessions for engineers: agentic coding workflows, practical model use, review loops, and guardrails. - Design workflows for safe AI-assisted development: context files, deterministic checks, repo conventions, build/test feedback. - Evaluate where LLMs are useful, where they fail, and how to structure the work so senior engineers retain control. ### Uber — Senior Software Engineer Worked on platform and payments systems at Uber scale. - Built internal server-driven UI infrastructure and developer-facing platform systems. - Worked on payments systems where reliability and cross-team ownership mattered at global scale. - Operated in large-codebase environments with build-system complexity, architecture review, and multi-team coordination. ### Independent / Open Source — AI products, tools, and consulting Built production tools and open-source systems around iOS, AI, developer tooling, and automation. - Built **WhisperBoard**, an on-device iOS transcription app: 50k+ downloads, 4.8 App Store rating, 1,035 GitHub stars. - Built **mcp-browser-use**, a Python MCP server for browser automation by AI agents: 938 GitHub stars. - Built **VibeSwitch**, an AI keyboard for rewriting text with sub-cent LLM call economics. - Built and operated local ComfyUI / diffusion pipelines for image and video generation, including Stable Diffusion-family workflows. - Mentor engineers through Codementor on iOS, AI integration, architecture, and practical debugging. ## Selected proof ### WhisperBoard Open-source iOS app for on-device voice transcription. - 50k+ downloads. - 4.8 App Store rating. - 1,035 GitHub stars. - Demonstrates mobile inference, privacy-first architecture, model packaging, App Store delivery, and real user traction. ### mcp-browser-use Open-source MCP server for browser automation by AI agents. - 938 GitHub stars. - Demonstrates early MCP ecosystem work, agent-facing tool design, browser automation, and Python infrastructure. ### Agentic iOS development workflow Six-plus months of agent-driven iOS development before Apple’s native Xcode agent support became mainstream. - Built preview-feedback tooling because agent loops fail without fast visual verification. - Structured repositories around explicit agent-readable conventions, build scripts, and review gates. - Practical thesis: the bottleneck is usually feedback-loop design, not prompt wording. ### ComfyUI / Stable Diffusion production workflows Hands-on work with local generative pipelines, including Stable Diffusion-family workflows and video generation. - Model management, GPU workflow debugging, failure handling, and reproducible local pipelines. - Focus on turning experimental workflows into repeatable systems rather than demo notebooks. ## Technology **Languages:** Swift, Python, TypeScript, JavaScript, PHP **iOS:** SwiftUI, UIKit, TCA, Tuist, App Store delivery, on-device ML **AI/LLM:** Anthropic, OpenAI, Claude Code, Cursor, MCP, structured output, evals, guardrails **Generative media:** ComfyUI, Stable Diffusion-family models, image/video generation pipelines **Backend/data:** FastAPI, Postgres/Supabase, SQLite, vector/RAG trade-offs **Infra:** Hetzner, Coolify, Vercel, CI/CD, build automation **Practices:** agentic development, code review systems, test harnesses, cost engineering, developer enablement ## Education / background Started programming at 16. Originally from Ukraine; based in Amsterdam. Long-running focus on building tools that compress the distance between idea, code, and shipped product. ## Links - Website: https://tarasenko.dev - GitHub: https://github.com/Saik0s - X: https://x.com/saik0s - Codementor: https://www.codementor.io/@saik0s