Careers

AI-native full-stack product engineer

See what you would do with us and apply in a few minutes. Feel free to send this page to someone it fits.

AI-native full-stack product engineer

One engineer who takes things from idea to production: frontend, backend, mobile, cloud, all built with AI. You work on the Abigel.ai voice platform, SaaS products, telecom/voice and web3 projects, use AI coding agents as everyday tools, and ship fast with real product sense.

Remote within Europe, or hybrid in Budapest Full-time or part-time; contractor is also an option

Details

What you will do

Build web front ends with React, Astro and TypeScript, from marketing pages to the customer admin panel
Write backends and APIs in Node with PostgreSQL and Redis
Develop mobile apps in Flutter or React Native and see them through app store release
Work on the Abigel.ai voice agents: speech recognition, LLM, speech synthesis, SIP and WebRTC calls, RAG and LLM API integrations (Gemini, Claude, OpenAI)
Use AI coding agents as everyday tools, and review their output the way you would review a colleague’s pull request
Write tests for whatever can break, and run what you build: Docker, CI/CD, Linux servers, monitoring
Join custom development, telecom and web3 projects when client work calls for it
Decide directly with the business side what gets built first and what can wait

Days and weeks

Every day: a short morning check-in on what ships today, then long, uninterrupted blocks of building
Every day: you review and test what the AI agent wrote before anything reaches the main branch
Mondays: you and the founder set the week’s priorities and agree what gets dropped
Every week: at least one release to production, with a short changelog
Every week: you go through anonymised call logs and find where the assistant got stuck
Every month: you look at cost and latency (LLM tokens, STT/TTS minutes, servers) and decide what is worth optimising

Your first 90 days

Day 30
You know the codebase, the servers and the release process. Within your first two weeks a fix or small feature of yours is live and in front of users.
Day 60
You take a full feature from plan to production on your own, with tests and monitoring. The Abigel.ai voice pipeline, from ring to reply, is no longer a black box to you.
Day 90
You own an area, such as calendar integrations or the mobile app. You propose what gets built there next quarter and put an estimate on it.

Who you are

You have shipped a production product end to end and can show it
You are confident in TypeScript and Node, and an SQL query doesn’t scare you
You use AI coding agents every day and know when not to trust them
You have built something on an LLM API and understand its limits (latency, cost, hallucination)
You write tests, and you don’t take an AI agent’s word that something is “done” until it has run
You are at home with Linux and Docker, and can work out from a production server’s logs what went wrong
You read and write technical English well

Nice to have

You have worked with real-time voice: telephony, SIP, WebRTC or streaming STT/TTS
You have a published Flutter or React Native app
You know the basics of web3 and blockchain
You have run a side project of your own that real people use

Tools

Claude Code, Cursor, Codex
TypeScript, Node, React, Astro
PostgreSQL, Redis
Docker, Linux, CI/CD
Flutter or React Native
Gemini, Claude and OpenAI APIs
STT/TTS, SIP, WebRTC
Git and automated tests

Who you work with

The founder: sets priorities, and you talk to them directly with no layer in between
Fellow engineers: code review both ways and shared architecture decisions
The key account manager: when a client asks for a custom integration, you say how much work it is and where the risk lies
Product marketing: before a launch you pin down together exactly what the new feature does and what it doesn’t

What we offer

What you build is live within days, and you see how people use it
Short decision paths: you take your questions straight to the founder
Flexible hours, remote within Europe or hybrid in Budapest
Access to the AI tools you need for the job
A live product with real users across the Hungarian, international and Spanish-speaking markets
Time and tools to keep learning

1. Test
2. Video
3. Task
4. Details

Short personality check Not a scientific test

Eight statements, answer on instinct. This is not a scientific psychometric test; it just helps us have a better conversation at the interview. There are no right or wrong answers.

Logic task

One short puzzle, solved with your own head. Write your answer and one sentence on how you got there.

Your answer

The water is rising, cats are stuck on the rooftops and you have one boat. The model (Gemini Flash-Lite) gets the engine description and a random level that you do not see in advance. Your job is a prompt that makes the model plan and run the rescue on its own. Typing commands or a pre-computed order will not help: every run uses a different level. You get five attempts; we measure which one succeeds and how many tokens you use.

The problem

The water rises 10 cm per time unit. When it reaches a cat’s height, that cat drowns.
The boat starts at column 0, the dock, and moves one column per time unit.
The boat holds one cat. Picking up and dropping off are instant; you can only drop off at the dock.
Every level has exactly one order that saves all five cats. The obvious strategies (lowest, nearest or farthest cat first) do not work.
Every run uses a new random level that only the model gets (at the end of the engine description, as LEVEL). After the run you see here which level it played.

Example level

Just so you can see what a level looks like. The model gets a different one on every run.

Cirmi: column 7, 150 cm up
Bodza: column 1, 350 cm up
Pötyi: column 2, 160 cm up
Mázli: column 4, 250 cm up
Füge: column 3, 310 cm up

A good prompt does not give the answer; it tells the model how to find it. The model may think in # comments before the program.

Engine description (the model gets this as its system instruction)

You control a rescue boat in a tiny simulation called "Flood". Reply with a program only: one command per line, no explanations, no markdown.

WORLD
- A flooded street with columns 0..8. Column 0 is the dock (dry land).
- The boat starts at column 0, empty, at time 0.
- The water level starts at 0 cm and rises 10 cm per time unit.
- Five cats sit at fixed heights in some of the columns. A cat drowns as soon as the water level reaches its height (level >= height).
- The cats of this run are listed in the LEVEL section at the end. Every run gets a new random level.

COMMANDS
move <col>   travel to column <col>; takes |col - current column| time units, the water keeps rising meanwhile
pickup       take the cat at the current column into the boat; instant; the boat holds one cat
drop         unload the cat at the dock (column 0 only); instant; the cat is safe
wait <n>     do nothing for n time units

RULES
- A command that breaks a rule stops the program: pickup with a full boat or where there is no cat, drop away from the dock or with an empty boat, an unknown command.
- Lines starting with # are comments; you may think in comments before the commands. At most 100 commands.
- The run succeeds when every cat has been dropped at the dock.

LEVEL
- Cirmi: column 7, height 150 cm
- Bodza: column 1, height 350 cm
- Pötyi: column 2, height 160 cm
- Mázli: column 4, height 250 cm
- Füge: column 3, height 310 cm

Your prompt

Before each step, please confirm you are not a robot.

AI-native full-stack product engineer