All vacancies
Vacancy

AI Product Engineer

Tashkent · Office · Full-time

Who we are

HUNTME builds products for logistics companies: Jobs — a job board for logistics professionals; HRMS — HR records, time tracking and payroll; Marketplace — a listings platform; Skills — online courses; and Fleet. The products share one tech stack and are moving onto a shared login and shared payments. The team is in Tashkent; our CEO works from the US.

How we work

AI writes our code — mostly Claude. A person understands the task, hands it to the AI, checks the result and owns what reaches clients. That's why we don't split people into backend, frontend and mobile: one engineer carries a change from start to finish.

The head of product decides what comes first. How to build it, and how to ship it without breaking anything for clients, is your call.

Our stack

  • Backend: Python, FastAPI, async SQLAlchemy 2, PostgreSQL, Alembic migrations, Redis, background jobs on PostgreSQL-based queues.
  • Web: React 19 and TypeScript, on Vite and Next.js.
  • Mobile and desktop apps: Flutter.
  • Shipping: Docker, GitHub Actions, AWS (ECS) for backends, Vercel for web. Tests: pytest.
  • AI in the products: Claude via the Anthropic API and Amazon Bedrock, Gemini in some products; semantic search on Qdrant.
  • Shared services: login, and payments via Click, Payme and Stripe.

You don't need to know all of this upfront: you'll pick up the unfamiliar parts with AI. What's required is listed under "Who we're looking for".

What you own

  • Changes in our products, from a client request to release: understand the problem, write the task, hand it to the AI, check it and ship it.
  • The product you lead — and a neighbouring one when needed. Our products share a stack and rules, so moving between them means learning a new domain, not a new job.
  • Bugs reported by clients: reproduce them, find the rule behind them and, if needed, clarify with the client together with our account manager.
  • Money: salaries, accruals and payments must add up exactly — no "roughly".
  • Data: one client company never sees another's data.
  • AI inside the products: assistants that answer users' questions about their own company and data.

Example. A client says they paid to promote a listing, but it didn't move up. You find where the rule differs from what the client was promised, write a test, have the AI fix it, check it on the test environment, ship it — and tell the account manager what to say to the client.

What success looks like after 90 days

  • You ship changes from request to production on your own, and every release has a test for the rule and a note of what was checked.
  • You know your product well enough that the team comes to you with questions about it.
  • You understand how it connects to the other products — through the shared login, payments and company data — and your releases don't break them.

Who we're looking for

  • You've built backends in Python — FastAPI, Django or similar — and were responsible for the code working for clients.
  • You're confident with PostgreSQL: you read and write SQL, understand indexes and transactions, and have changed the schema of a database with live data without clients noticing.
  • You write tests — pytest or similar — and know that a test checks a rule rather than repeating the code. You work through Git and pull requests, run a service in Docker and find in the logs where and why it failed.
  • You work with AI every day — Claude Code, Cursor or similar — and can show on screen what you gave the AI, where it went wrong and how you noticed.
  • You've worked with money or accounting: payroll, accruals, payments. You know why you can't round numbers however they come out.
  • You can talk to a user — a dispatcher, an HR officer, an accountant — and understand what they actually need.
  • You write clearly: a task, a decision, a short release note. Our CEO is in the US, so we decide a lot in writing.
  • For the React frontend, it's enough to read and edit it together with AI — deep experience isn't required.

What you catch in AI-written code

AI writes fast and makes mistakes exactly where they cost the most. Catching them before release is your job. For example:

  • Permissions or the company are checked only on the frontend — so a client who puts someone else's ID into a request sees another company's data.
  • Money is calculated in floating-point numbers or rounded at the wrong step — and the payroll total doesn't match the bank statement.
  • The payment provider sends the same notification twice — and the payment is counted twice.
  • A migration on a large table locks the database for minutes or loses old data.
  • A database query inside a loop: fast on test data, half a minute per page on real data.
  • Time is stored without a time zone: the server runs in UTC, but the company's working day follows its own time zone.
  • A test is always green because it checks nothing.

Nice to have

  • Flutter mobile apps — at least reading and editing them with AI.
  • Integrations with external systems: payment services (Click, Payme, Stripe), time-tracking terminals, government services.
  • You've built APIs for AI agents or MCP servers (a way to give an AI agent access to a product), semantic search on a vector database (Qdrant or similar), or evaluated the quality of AI answers.
  • You know AWS and GitHub Actions well enough to work out on your own where and how a service runs after release.
  • Experience in logistics or with HR systems.

This role is not for you if

  • You prefer to get a finished spec and hand your work to others to check. Here you decide how to build it, and you own the release.
  • You want to stay in one part of the system for years. Here you go where you're needed most.
  • You don't use AI in your work and aren't ready to start on day one.

How hiring works

  1. A call, 15–20 minutes: we get to know each other and talk about expectations and terms.
  2. An experience interview, 45–60 minutes: your projects, production incidents and how you fixed them, plus technical questions — money, client data, changing a live database.
  3. A live practical session, 60–90 minutes: a training project on our stack — FastAPI, SQLAlchemy and Alembic. You work with your own AI tool — that's required. We look not at speed but at how you set the task, check the result and what you notice yourself.
  4. A final conversation. The hiring decision is made by a person.

Terms

  • Full-time.
  • Office in Tashkent.
  • Probation: 3 months on full terms. We review it against the results listed above under "What success looks like after 90 days".
  • No 24/7 on-call. Outside working hours we only call you for a critical incident — money, login or client data; those hours are paid under Uzbekistan's Labor Code.
  • Claude Team Premium paid by the company.

How to apply

Message @huntme_hindol on Telegram. In your message, describe in two or three sentences the last task you did with AI: what worked and where the AI got it wrong. If you have code you can share, add a link.

Ready to apply?

Message @huntme_hindol on Telegram.