Most AI companies fall into one of two traps. Either they promise to AI-ify everything and ship
something nobody can depend on, or they build solid software and bolt an AI feature on top as a
sales hook.
We do neither. We build automation systems for real operations, where the software owns the control flow and AI is used only where nothing else works: messy documents, unstructured input, judgment at a volume humans can't sustain.
Every AI step in our systems has a defined contract, a measured error rate, and somewhere
to go when it isn't sure. If we can't measure it, we don't ship it.
What you'll do
- Build and own production automation systems end to end, deployed inside client operations
- Work with the client's actual mess: ERPs, CRMs and industry platforms never built to talk to each other, vendor systems with no real API, undocumented databases, scanned documents, Arabic and English in the same file, and the exports and email attachments people use to bridge the gaps
- Decide where deterministic code handles something and where a model has to, then prove the model is right often enough to trust
- Build the evaluation harnesses that tell us our error rate on real data, not on a benchmark
- Keep systems running after launch. Monitor, catch drift, retune thresholds, handle the exception queue
- Sit with operations teams, watch how they actually work, and come back with something that survives contact with them
What we're looking for
- 3+ years building and shipping software that real users or a real business depended on
- You can sit with a client's operations team, work out how their business actually runs, and come back with a technical design. This is the part we care about most
- Strong Python or TypeScript, and you have built and deployed real services: APIs, background jobs, queues, scheduled work, retries and idempotency. Not just scripts
- Solid SQL and schema design. You can model a client's process into tables and query it without help
- Real integration experience: third-party APIs, OAuth and API key auth, webhooks, pagination, rate limits, and the failure handling that comes with all of them
- You have worked with unstructured input: PDFs, scanned documents, OCR, parsing email and messages
- You can deploy and run what you build: Docker, a cloud provider, CI, logs and alerts. When something breaks in production, you can find the cause and fix itYou have used LLM APIs in production, including structured outputs and function calling
- You can own a piece of client work end to end without being managed through it, and you know when to escalate instead of guessing
- You write clearly. Most of the work is remote, so specs, updates and handover notes carry a lot of weight
- Previous consulting or client-facing engineering experience is a plus
How we work
Mostly remote.
On-site is for the parts that don't work any other way: discovery sessions at the client, go-live, and the days when something is broken in a way that needs you in the room. Expect a few days a month, clustered around those moments rather than spread evenly. We'll also get together in person regularly as a team.
You'll have a lot of room. We agree what needs to happen and by when, then you get on with it. No daily check-ins, no tracker to feed. We review the things that matter and help whenever you ask. The team is young and growing, so what you own expands as fast as you want it to.
What this is not
- Not a research role. We don't train models
- Not a prompt engineering role. Prompts are a small part of a much larger system
- Not an agent-building role. Agents are rarely our first choice, and the job is much wider than that
- Not a role where you hand over a prototype and move on. You own what you build
Additional Details
- Start: Sooner is better
- Language: Arabic is a strong plus, not a requirement
- Process: Intro call + Technical conversation
- Benefits: Health insurance, Paid Annual leave and Ticket.
