AI Systems Engineer

WHAT THIS IS

We build AI employees for small businesses: voice and WhatsApp agents that

answer the phone, book the appointment, follow up, and write it all back into

the customer's records. Real companies, real phones ringing. When something

breaks at nine in the morning, a restaurant stops taking bookings.

It all runs on Linux VM. Postgres. nginx in front of everything.

That is a deliberate choice, and it means one person can hold the whole system

in their head. It also means there is nowhere to hide a mistake.

We are not hiring someone to type code. The machine types now, and it is good

at it. We are hiring the judgment that decides what gets built, knows what a

change breaks three layers away, and can tell a result that is correct from one

that is merely convincing. You will use AI all day. You will also be

accountable for everything it produces.

THE WORK, IN THREE PARTS

1. The machine it runs on. Docker Compose, nginx, TLS, one Linux VM. Deploys,

certificates, and backups you have actually restored from.

2. Integrations against suppliers who fail. Booking systems, telephony,

messaging, invoicing. They rate-limit you, they time out, they return a 200

with nothing inside it, and once in a while they change the shape of their

answer without telling anybody. Most of this job is deciding what our software

does in that second.

3. LLM plumbing. Tool calling, structured output, and the interesting part:

what the code does when the model returns garbage. Not prompt writing. The

scaffolding, the guards, and the path the system takes when the model is

confidently wrong.

WHAT YOU NEED TO KNOW ALREADY

SQL, on Postgres 16. Not an ORM you trust. The query, the plan, the lock, the

migration, and the way back out of it.

JavaScript, on Node 20. Async, failure, timeouts, retries, and what happens

when two things run at once.

That is the whole list. No degree, no minimum number of years, no framework

checklist. Everything specific to us is learnable in a fortnight.

HOW YOU WORK, WHICH IS THE REAL FILTER

You go underneath the description you were handed, and you can then say which

limits are genuinely fixed and which are just a choice somebody made.

You explain complicated things in plain words, to people who do not share your

vocabulary.

You judge work by how it is put together, where it gives way and what it costs,

not line by line. The machine does not make typos any more.

You can name how something breaks before it breaks, and under what conditions.

You never make a change without naming the way back first.

On the second occurrence you fix the process, not just the instance, and you

leave a rule or a check behind.

You use AI as a matter of course, you supervise it because that is the job, and

you read fast enough to catch it being confidently wrong.

USEFUL, NOT REQUIRED

Agent-to-agent protocols such as MCP or A2A, or building agents that other AI platforms can

call. Analytics dashboards or CRM-style products. Knowing what it takes to pass an enterprise vendor's due diligence.