Stepan Kotok – AI Engineer, Full Stack Developer, Solution Architect

Profile

In software development since 2014; since 2020 led projects as a tech lead at my own company, Neatapps, for clients in the US and Europe. Since 2024 I have been systematically strengthening the whole engineering cycle with AI-driven development.

Co-founded Neatapps with a partner in 2015. First I wrote code, then led products end to end as tech lead and Solution Architect: from client alignment and requirements gathering to architecture, development, testing, release and support. Clients were mostly from the US and European countries.

In March 2026 I sold my share of Neatapps to focus fully on AI development. Now, with AI agents, I handle tasks that used to require a team of 3–4 people led by a manager.

Team role

Primary role: Senior developer responsible for technical reliability. I choose technologies for the task, design the system architecture, write code and control quality at every stage.
When needed, I take the team lead role: at Neatapps I coordinated work with clients and the team, and ran daily stand-ups with reports on completed work and a review of difficulties.
First I clarify what result is needed and how to verify it, then I plan, align and start the work.
I prioritize tasks by business value and technical feasibility according to approved plans, deliver them step by step in short development-and-testing iterations, and release to production.

How I write code with AI

Tools

Claude Code: – main tool, used daily, on the top Max plan. Subagents, skills, hooks, custom MCP servers, parallel work in isolated git worktrees. I choose the model for the task: a stronger one plans and accepts the work, a faster one writes code from a brief with acceptance criteria.
Cursor: – proficient; used on a paid subscription; my main tool is now Claude Code.
ChatGPT/Codex, Gemini, Grok: – on the paid Plus, Pro and Super plans: independent review of code and plans. Several models in turn, with fixes between rounds; a second opinion before irreversible steps: a migration, a deployment, a change of access rights.
Perplexity, NotebookLM: – research: survey of recent approaches, documentation, comparison of solutions, collecting and cross-checking sources.
Browser and phone testing (Playwright, Mobile-MCP): – agents verify the result themselves in a real browser and on a mobile phone: end-to-end scenarios, accessibility, screenshots.

How this speeds up work without lowering quality

Faster code. In four days I built, with AI agents, an online course platform with offline mode, sign-in and access rights, and an admin panel, covered by end-to-end automated tests.
Faster bug finding. I hand log and error analysis to agents: they find the point of failure and propose a fix, and I check the diagnosis and accept the result.
Faster, more thorough testing. Agents write automated tests in code and scenarios for manual testing. On the AI assistant project for a tourism service (below), I wrote 1,310 tests in six layers in seven days. The AI also tests like a live tester: it runs scenarios in a real browser and on a mobile phone, takes screenshots and reports what to fix.
Faster, safer releases. A blocking check before the main branch, every change linked to a task, CI in GitHub Actions.
Routine goes to agents. Repetitive edits across dozens of files, issue triage and documentation search are done by agents. I handle architecture, difficult parts and verification of the result.
I verify rather than trust. When the AI makes a mistake, I record a rule for it so the mistake does not repeat. Any generated code is first verified with typing, tests and architecture review, and only then adopted.

Selected projects

AI assistant for a tourism service: hotels, routes, booking

2026 · Next.js, TypeScript, Supabase/PostgreSQL, Claude API and OpenAI API, Vitest, Playwright, GitHub Actions

Willy Agency project, Saas-Fee, Switzerland · Tourism service

The owner built the product with AI agents in 71 days; it was in operation and close to its first sales. In one week in August 2026, directing AI agents, I:

Set up reliable operation of the conversational assistant with the tourism service's knowledge base (hotels, routes, guests' service questions, slot booking).
Built tests from scratch: 1,310 tests in six layers, with no network, database or model; a full run takes 3.64 seconds and needs no secrets.
Added a blocking check before the main branch and linked changes to tasks: 62 of 63 merges instead of 24 of 107; CI went from nothing to four pipelines.
Audited the whole surface: route authorization, row-level data isolation (Supabase RLS to protect private data from prompt injection), incoming webhooks, secrets, environment separation; closed the critical issues and locked them in with tests that fail the build if a problem returns.
Moved the emergency scenario out of the model's control: a fixed reply with rescue service numbers fires before the rate limiter.
Expanded the written rules for AI agents from 14 to 339 lines, each added after a measured failure.

Device condition assessment app

Neatapps · NDA project for an international consumer electronics company · Flutter (iOS, Android), native iOS, Node.js, computer vision

Turnkey: interface design, a Flutter mobile app with native iOS modules, and a Node.js backend. Real-time computer vision detected the device's position in front of the camera and took photos itself, and an API returned the condition assessment from them.

MCP server for Kommo/amoCRM

2026 · own project · TypeScript, MCP SDK, Cloudflare Workers, OAuth, zod, Vitest · code available on request at interview

Gives an AI assistant read-only access to a CRM: more than 20 tools covering deals, contacts, pipelines, tasks and lead-source performance. Data is minimized: click and visit identifiers are stripped; CRM content is passed to the model as data, not as instructions; output is size-limited and states plainly when an answer is incomplete. The local version and the remote one on Cloudflare Workers are ready; I am now adding OAuth.

A system of AI agents around Claude Code

2026 · own project · Claude Code (subagents, skills, hooks), Bash, Python, Codex CLI, Grok CLI, Antigravity · code available on request at interview

A strong model plans and accepts the work, Sonnet subagents execute it, and mechanical file work goes to Grok and Gemini in a macOS sandbox with an action-log check. Before irreversible steps (deployment, migration, history rewrite), the system asks Gemini, Grok and ChatGPT in turn: the advice can only stop a step, and a measurement is what allows it. Token usage is counted by custom scripts.

transcribe-hard – transcription of difficult recordings

2026 · own project · Python, MLX, Parakeet, Whisper large-v3, GigaAM, Grok STT, whisperkit, ffmpeg · code available on request at interview

For recordings with noise, a distant microphone and interruptions: several recognition models, word-by-word comparison, disputed spots, speaker separation. The AI agent resolves only the disputed parts; the code assembles the result and checks that no word is lost or invented. One hour of audio takes about 18 minutes on a MacBook Air M3; 48 tests.

Own products

2025–2026 · web and Android

Pearls – an online course platform: a PWA with offline mode (Service Worker, IndexedDB, TanStack Query with persistence), Supabase with RLS policies and pgTAP database tests, end-to-end tests in Playwright, Cloudflare Workers.
MathPaw – a multiplication table trainer: offline storage in IndexedDB (Dexie), Supabase, error monitoring in Sentry, Vitest and Testing Library, Netlify.
AI Voice Rec – an Android voice recorder: transcription on the phone itself without internet (Whisper and SenseVoice via sherpa-onnx), titles and summaries via Gemini. Kotlin, Jetpack Compose, Room. Code available on request at interview.

Work experience

Willy Agency – technical partner (contract)

August 2026 – present · part-time · Saas-Fee, Switzerland, remote
- Responsible for the engineering side of an AI assistant for a tourism business: tests, blocking pre-release checks, security audit, rules for AI agents (details in "Selected projects").
Full Stack / AI Developer (contract)

April 2026 – present · Chișinău
- Take AI-assisted prototypes to production: architecture, data, authorization, payments, tests of critical scenarios, release.
- Connect AI to business data: LLMs on top of company knowledge, RAG, MCP, process automation.
- Lead a product end to end: from requirements, design and architecture to release and support.
Neatapps – Co-founder, Technical Project Lead

July 2020 – March 2026 · clients in the US and Europe
- Turned client goals into a technical scope of work: research, requirements, estimates, delivery plan, client communication.
- Coordinated developers and designers across the whole cycle: architecture, development, testing, release, post-release support.
- Owned technical decisions and risks; moved the company from one-off orders to full-cycle product development.
- NDA projects, without names:
  - a device condition assessment product for an international consumer electronics company: Flutter (iOS, Android), Node.js backend, native camera integration, real-time on-device computer vision;
  - a mobile emergency care reference app for a medical organization;
  - tools for IoT devices of an electronics manufacturer.
- In March 2026 I sold my share in the company to work more closely with AI.
Neatapps – Co-founder, Mobile Developer

July 2015 – July 2020
- Native Android (Java, Kotlin), then iOS (Swift) and cross-platform development in Flutter.
- Architecture, APIs, integrations, including BLE devices, testing, publishing to the App Store and Google Play, support.
- Worked directly with clients: clarified requirements and translated them into a technical solution.
Irpin Bible Seminary – System Administrator

February 2004 – September 2014
- The organization's computers, network and internet, server roaming profiles for staff and students, user support.

Skills

Frontend:: React, Next.js, TypeScript, Vite, Tailwind CSS, PWA (Service Worker, IndexedDB), TanStack Query.
Backend and data:: Node.js, NestJS, RESTful API, PostgreSQL, Supabase (RLS, RPC, migrations, pgvector), pgTAP.
AI and LLM:: Claude API (Anthropic SDK), OpenAI API, Gemini API, MCP (custom servers), RAG, Function Calling, structured outputs.
Architecture:: Solution Architecture, System Design, data security.
Cloud and DevOps:: Docker, Cloudflare Workers/Pages, Vercel, Netlify, GitHub Actions (CI/CD), Git, Linux.
Mobile development:: Flutter, Android (Java, Kotlin), iOS (Swift), BLE/IoT.
Quality and reliability:: Vitest, Testing Library, Playwright, pgTAP, axe, Sentry.
Used previously / familiar:: Java (Core, Spring), Python (FastAPI, SQLAlchemy), Redis, MongoDB, AWS, GCP, Nginx.

Languages

Russian, Ukrainian: – native.
English: – B2, conversational: calls, correspondence and documentation with clients from the US and Europe.