Engineering · 120 min · AI allowed

Senior Backend Engineer, AI Systems

Two hours as the engineer who owns search, ranking and AI at a stays marketplace: build hybrid search with live prices in a private GitHub repository and measure it, design the answer to a guest's sentence, clean a messy business feed, personalise with consent, design a language-model feature that can't be hijacked, and handle an assistant that has been. AI tools welcome; we score your decisions, then your craft.

Open for applicationsStart when you're readySame criteria for everyone
01 / What you'll do

5 short tasks.

01

Build hybrid search

In your own private repository: turn a guest's sentence into hard rules and meaning, keep every rule, take prices live, keep each business's data separate, and measure it with a test set you build. We run it against hidden queries.

02

Design

Walk through how search should answer a real guest sentence, and design an "Ask about this stay" feature with a language model: grounding, live data through tools, prompt injection, personal data.

03

Data and personalisation

Clean a messy business feed and say how you keep it clean and prove it; rank for a returning guest with consent, and explain why each stay is there.

04

Handle the day

Questions from product, partners, privacy and the CTO arrive while you work, and the AI assistant goes wrong in production partway through.

05

Your own system

Walk us through an AI system you built and ran in production, with numbers.

02 / What we look for

How we score this role.

Everyone applying for this role is scored against these same criteria. Where possible, automated checks test whether your work really holds up; people review your reasoning.

01

Hard rules and meaning

Place, dates, guests, pets and price always hold; meaning orders what they allow. You know when not to use vectors.

02

Truth where it lives

Prices and availability come live from the system that owns them, never from a copy.

03

Measured

A test set, metrics and an honest before and after, not a demo.

04

Separate and safe

Each business sees only its own data; host text can't steer a model; personal data stays out of prompts and logs.

05

Production experience

Systems you ran at real usage, what broke and what you changed. Using AI coding assistants is expected, and it is not this experience.

06

Evidence over claims

Evaluation reports, dashboards and code count more than titles. "Not yet" is a fine, honest answer.

03 / How it works

From invite to call.

1

Open your invite

Use the personal link in your email from careers@automaark.com. No account or password needed.

2

Show us your work

Add your own accounts, accounts you've worked on and your portfolio, with a screenshot of each. Not timed.

3

Press Start

120 minutes in your browser, in one sitting. Everything saves as you type.

4

Submit

Or let the clock run out. We review what you saved.

What you'll need
  • A laptop with Node 22 (or Python 3.12 if you prefer to rewrite the starter), Git and your editor
  • A GitHub account
  • Links to systems you built (search, retrieval, ranking, LLM features), and screenshots of real work if you can: an evaluation report, a dashboard, a design doc (redacted)
  • Your usual AI tools
  • Two uninterrupted hours
The rules
  • Any AI tool, docs or search. No help from another person.
  • You own every part of what you submit, whoever or whatever wrote it. Check it.
  • Tell us where AI helped and where it was wrong.
  • Partial work counts.

Read how we use AI at Automaark →

04 / Questions

About this assessment

Do I need GitHub?

Yes. Before you start, you connect your GitHub account and we create a private repository with the starter, the data and a local copy of the live price service. It opens to you when you press Start and closes when your time ends.

Which language?

The starter is TypeScript with NestJS, our stack. You may rewrite it in Python or anything else, as long as the two scripts and the API contract in the README stay.

Is there a language model I can call from my code?

No: the search is graded offline. The repository includes a small offline text embedding you can use, and you design the language-model feature in the browser. Use any AI tools you like while you work.

How is the repository graded?

We run your search against hidden queries with our own live prices and check every hard rule, that no business sees another's private stays, that prices are live, ranking and speed, then run your own evaluation. People read your notes, evaluation and code.

Is this really from Automaark?

Yes. Labs runs on automaark.com, our own domain, and invitations come from careers@automaark.com. We will never ask for a password, payment or ID documents. If in doubt, email careers@automaark.com.

Can I use ChatGPT, Claude, Copilot or Cursor?

Yes, and we expect you to. You own everything you submit, so check what AI gives you, and tell us where it helped and where it was wrong. No help from another person. The full rules are on our AI-use page (labs.automaark.com/ai-use).

When does the clock start?

Only when you press Start. The deadline in your email is the last moment you can start. Pick any time before then that suits you; once you start, you finish in one sitting.

What if something breaks on your side?

If our platform fails during your attempt, it's on us. Reply to your invite email and we'll give you a fresh attempt.

What happens after I submit?

Your work is scored against the same written criteria as everyone else applying for that role. Shortlisted candidates are invited to a 30-minute call about what they did.

What do you do with my data?

We store your name, email, the links and screenshots you give us and your assessment work, only for this hiring process. Your work stays private, is never published, and is deleted within 12 months.