Yaroslav Salimov
DesignStatus page
CV

From client idea to reliable, tested production.

AI Engineer and Forward Deployed Engineer. I work out what the client actually needs, then build it end to end: backend, ERP and CRM integrations, LLM features, tests and the release. Nearly 5 years with Node.js and TypeScript.

  • Kraków, Poland · UTC+1/+2
  • Remote
  • No sponsorship needed

Latest update · Intro · 2 min · AI-generated video

Hi, I'm Yaroslav. My last three projects in two minutes.

0:00 / 2:24
Transcript

Hi, I'm Yaroslav. I'm a full-stack and AI engineer, and for almost five years I've been building products with Node.js and TypeScript. Let me tell you about my last three projects.

Right now I'm at RepMove, a CRM for equipment dealers. I connect the CRM to our customers' systems: two-way integrations with ERPs like NetSuite, Epicor and Texada, and data migrations from Salesforce and Microsoft Dynamics. I tracked down silent data loss in production: almost 6,000 deals had gone missing, and I found the cause and planned the resync that brought them back. I also built an MCP server, so AI assistants like Claude can work with a customer's sales data. This summer I merged 110 pull requests with zero reverts.

I've also spent over two years on Thinkable, a mental-health platform for clinics. I was the main developer of the clinical dashboard and the backend for three mobile apps. I shipped LLM features with OpenAI and Gemini, like a safety screen for patients' messages. I cut our slowest AI feature from 57 seconds to 5, at a tenth of the cost, and I built the partner API for a clinic pilot in three weeks.

From 2023 to 2025, I built TaxZap, a tax-refund service in Ireland and the UK, end to end: the backend, the client web apps, the admin panel and the infrastructure. That covers payments, the tax calculation engine, and a no-code questionnaire builder that runs the tax-return intake in both countries.

I work AI-native every day, with Claude Code and Codex. That's why a full two-way CRM integration takes me about a week.

I'm based in Kraków, open to remote AI engineer, forward-deployed and full-stack roles, and I can start right away. If you need someone who takes an idea all the way to production, I'd be glad to talk.

Key metrics

  • 3 weeksto build a clinic pilot’s partner API; the first clinic went live in September 2026
  • 57 s → 5 sLLM response time in production, at about 10× lower cost per call
  • 110 PRsmerged in one summer, mostly on the hardest tickets, with zero reverts
  • 9 hoursfrom the first commit of my app Modbook to App Store review; now on Android too

Runbook

How I work

What happens between the idea and the release.

  1. Step 1

    Understand

    Work out with the client what they actually need, map the systems and data it touches, agree on what done means.

    A TMS-clinic pilot (TMS is magnetic brain stimulation for depression): payers’ eligibility rules became a rules engine for the screening, with patient stages and clinic handoffs. Built in 3 weeks; the first clinic went live in September 2026.

  2. Step 2

    Build end to end

    Backend, integrations, LLM features and UI, in small pull requests that get reviewed.

    The backend and React admin dashboard of a mental-health platform, from scratch: ~255 documented API routes.

  3. Step 3

    Prove it works

    Automated tests, dry runs on the real data and load tests before anything goes live.

    ~118 automated tests on the LLM safety screen for patients’ free text, a Playwright API test framework built in 6 days, k6 load tests for 197 functions.

  4. Step 4

    Ship and stay on it

    Release, watch production, and fix the root cause when something breaks.

    Found silent data loss in production syncs and planned the resync that restored 5,925 deals; found the root cause of a production outage with Cloud Logging and PromQL.

Components

Work

What I built, where it runs and its state today.

01 · Integrations & migrations · RepMove

175,000 accounts in. 14,000 wrong merges caught on the way.

In production
  1. 01Two sources for one equipment dealer: Salesforce and Oracle.
  2. 02Matching keys scoped by record type stopped 14k wrong merges.
  3. 03An import that is safe to re-run and starts with a dry run.

ERP & CRM

Two-way syncs with customers’ systems

At RepMove, a CRM for equipment dealers, I build two-way syncs with customers’ ERP systems. Texada SRM, multi-tenant and built from scratch: customers, contacts, quotes, reservations and contracts become deals, and changes are pushed back. Epicor P21 is designed for a 154k-customer backfill. I also extended the NetSuite, Point of Rental, QuickBooks and Integration.app syncs. All of it runs on Cloud Scheduler and Cloud Tasks.

  • Texada SRM
  • Epicor P21
  • NetSuite
  • Point of Rental
  • QuickBooks
  • Integration.app

Customer onboarding

Enterprise migrations

47k accounts, 72k contacts and 70k opportunities from Microsoft Dynamics; 57k clients for another dealer; 8 tenants merged into 1. Every import runs dry first and is safe to re-run.

02 · LLM features in production · Thinkable

57 seconds down to 5. About 10× cheaper per call.

In production

At Thinkable, a mental-health platform, OpenAI and Gemini features reach clinicians and patients. Personalization took 57 seconds. I made the AI service faster and cheaper with one config that picks the model tier per task, parallel batches and OpenAI’s priority tier, and I added AI usage tracking per route.

  • One model-tier config
  • Parallel batches
  • OpenAI priority tier
  • AI usage tracking per route

03 · Product frontend · Thinkable

The clinic side of a TMS pilot, and the editor behind the content.

Pilot

For a TMS-clinic pilot I built the clinic’s workspace: the patient list, a phone screening that applies each payer’s rules, documents and sign-off, and the pilot’s numbers. The screens use synthetic patients, so no real patient data is shown. In the admin dashboard I built from scratch, the content editor holds the app’s 12,795 statements in 13 languages. React 19, TypeScript and Vite for the workspace; React 18, Ant Design and RTK Query for the dashboard.

Clinic workspace: the patient list with stages and next actions

Every patient in one list. Change a stage in one click and the next action updates itself.

04 · MCP & AI-native delivery · RepMove

I build the tools agents use, and I ship with agents every day.

In use

Model Context Protocol

An MCP server for a public reporting API

  • Stateless Streamable HTTP on Cloud Functions
  • Tools reuse the REST endpoints’ Zod schemas and OpenAPI docs
  • Access limited per company

Used with Claude, it turned 66k accounts into a 12.4k-row address-quality report for a customer.

Claude Code · OpenAI Codex

AI-native, with guardrails

  • Wrote the project context the agents run on: CLAUDE.md and AGENTS.md, ten per-integration docs and an integration blueprint.
  • Turned ~340 AI code-review findings into coding guidelines and a pre-PR checklist.
  • Built a Playwright API test framework in 6 days: Claude subagents read the backend routes and Zod schemas, plan the tests and write them.

With this setup, a full two-way CRM integration takes me about a week.

05 · Apps I ship on my own · My own apps

My own apps for iOS and Android, from idea to the stores, built solo.

Live on iOS and Android

iOS · Jul 2026 · Android · Nov 2026

Modbook · Car Build Tracker

Built solo with Claude Code: v1.0 went to App Store review 9 hours after the first commit, and it is now at 1.5 with iPad support. Swift 6, SwiftUI, SwiftData mirrored to CloudKit, ~120 tests and CI on GitHub Actions. An in-app LLM prompt turns free-form notes into a clean CSV import. The native Android version followed: Kotlin and Jetpack Compose, 430+ tests, in Google Play review two days after its first line of code.

0:00

iOS · App Store · Sep 2025

TickDown · Countdown timer

A countdown app with Home and Lock Screen widgets, a Live Activity, Siri shortcuts and iCloud sync. I rewrote it as 2.0 in one day while keeping existing users’ widgets and data.

0:00

06 · Video ads · For my apps and clients

I also make video ads, for my apps and for clients.

Taking orders
0:00

Modbook has no marketing team, so I became its ad agency: story-driven vertical ads for TikTok, Instagram Reels and YouTube Shorts, from the script to the final cut, each in a new style.

I make the same kind of ads for other apps and startups.

Incident history

Resolved incidents

Problems I found in production, and what fixed them.

  • Resolved

    Deals disappearing from production syncs

    RepMove

    Production syncs were losing data without any error. I traced 5,925 missing deals to a cascade delete and planned the resync that restored them.

  • Resolved

    Contacts merged into the wrong records

    RepMove

    An external-ID collision had merged ~39k contacts into ~420 records, and pagination moved its cursor on after a partial fetch. I found both.

  • Resolved

    Slow cold starts

    RepMove

    About 2,000 cold starts a day at ~10 s each. My esbuild bundling prototype makes functions load 2.6× faster; it is now the team’s bundling spec.

  • Resolved

    A production outage

    RepMove

    I found the root cause with Cloud Logging and PromQL.

Uptime

Experience

Nearly five years of taking products from idea to production.

Months on the project, Nov 2021 to now

  1. RepMove · field-sales CRM SaaS

    Feb 2026 – now · Remote

    Full Stack Developer, integrations & AI

    Operational
    • Multi-tenant CRM for equipment dealers: about 200 Firebase Cloud Functions on Firestore and MongoDB Atlas. In Jun–Aug 2026: 110 merged PRs with zero reverts, mostly on the team’s hardest tickets.
    • Two-way syncs with customers’ ERPs on Cloud Scheduler and Cloud Tasks: Texada SRM from scratch, Epicor P21 designed for a 154k-customer backfill, Integrated Rental, Dodge and ConstructConnect; extended NetSuite, Point of Rental, QuickBooks and Integration.app.
    • Onboarded enterprise customers with imports that run dry first and are safe to re-run: up to 175k accounts and 181k contacts per customer, and 8 tenants merged into 1.
    • An MCP server for the public reporting API. Measured ~2,000 cold starts a day at ~10 s each and prototyped esbuild bundling that makes functions load 2.6× faster; it is now the team’s bundling spec.
  2. GGtude · Thinkable, a mental-health platform

    Jun 2024 – Oct 2026 · Remote

    Full Stack Developer · part-time, full-time Oct 2025 – Jan 2026

    Completed
    • Main developer of the clinical dashboard and the backend for three iOS/Android apps, with 495 releases.
    • Cut personalization time from 57 s to about 5 s. LLM features in production: a safety screen with crisis alerts, recommendations, AI progress summaries and a 13-language translation pipeline.
    • A partner API for a TMS-clinic pilot in 3 weeks; the first clinic went live in September 2026.
    • The multi-tenant backend and React admin dashboard from scratch: ~255 documented API routes, PHQ/GAD clinical reporting, English/Hebrew RTL.
  3. ArtiD · client TaxZap, tax refunds in Ireland and the UK

    Mar 2023 – Sep 2025 · Kyiv / Remote

    Full Stack MERN Developer

    Completed
    • Backend from the first commit: Node.js, Express and TypeScript on MongoDB with role- and region-based access, plus work across the admin and client apps.
    • A no-code questionnaire builder on React Flow and the client engine that renders it; it powers the Irish and UK tax-return intake.
    • A Revolut Checkout paywall, invoice and agreement PDFs, and an income tax calculation engine covered by Jest tests.
  4. DevPuzzle · web apps for early-stage startups

    Nov 2021 – Mar 2023 · Kyiv

    Full Stack Developer

    Completed
    • REST APIs on MongoDB and PostgreSQL with React front ends, including a scheduling and booking SaaS with Google Calendar and Zoom.
    • Authentication, access control, WebSocket real-time features and timezone-aware scheduling.

Bachelor’s degree in International Business, National Aviation University, Kyiv · English B2 (Upper-Intermediate) · Ukrainian and Russian native

Dependencies

Stack

What I work with.

LLM & AI
OpenAI Responses API, Google Gemini, MCP servers, Structured JSON output, Safety screening, Model tiering, Latency & cost tuning, AI usage accounting
AI-native development
Claude Code (subagents, skills, memory), OpenAI Codex, Agent-generated test suites, AI review → coding guidelines
Customer integrations
NetSuite (SuiteQL), Epicor Prophet 21, Texada SRM, Point of Rental, QuickBooks, Dodge, ConstructConnect, Integration.app, Zapier, Stripe, Revolut, Salesforce & Dynamics migrations
Backend
Node.js, TypeScript, Express, REST, OpenAPI (Scalar), Zod, WebSockets, JWT
Cloud & data
Google Cloud: Functions, Run, Tasks, Pub/Sub, Scheduler, Cloud Logging, PromQL, Firebase, AWS (EC2, S3, SES, SNS), MongoDB Atlas, Firestore, PostgreSQL, MySQL
Frontend & mobile
React 18/19, Redux Toolkit, RTK Query, Ant Design, React Flow, Flutter, Swift & SwiftUI, Kotlin & Jetpack Compose
Quality & delivery
Jest, Playwright, k6, esbuild, Docker, GitHub Actions, nginx, pm2

Contact · Available immediately

Open to remote roles.

I’m looking for AI Engineer, Forward Deployed Engineer or full-stack TypeScript roles. Kraków, Poland (UTC+1/+2), permanent residence, no sponsorship needed.