Lead AI Engineer · Python & Agentic Systems
Lead agentic system design in Python across client engagements — architecture, evals, and the production behaviour of agents that clients depend on.
What this seat actually is.
We are a Claude-native studio: most of what we ship is built on Claude and Claude Code, and agents are the centre of the work rather than a feature bolted onto it. This role leads that engineering.
You own the architecture of agentic systems from the first whiteboard through production — how work is decomposed into tools, where state lives, what the model is allowed to do unsupervised, and how anyone can tell whether the thing is getting better or worse. Solution architecture is half this job: you sit with clients, take an ambiguous business problem, and come back with a design that a pod can build and that operations can run.
Demos are not the deliverable. An agent that impresses in a notebook and quietly degrades in month three is a failure. We ship with evals from day one and we stay on the pager afterwards.
Fully remote from the UAE, async-first. We work in writing.
What you own.
- ·Own the architecture of agentic systems end to end: tool boundaries, orchestration and hand-off, memory and state, retries, guardrails, and the human-in-the-loop points.
- ·Lead solution architecture with clients — turn an ambiguous business problem into a written design with explicit trade-offs, failure modes, integration surface, and cost model.
- ·Build in Python to a production standard: typed, tested, observable services, not scripts promoted by accident.
- ·Design and maintain the eval harness for every system we ship — golden sets, regression suites, and the offline and online signals that tell you a change actually helped.
- ·Own retrieval where it is warranted: chunking and indexing strategy, hybrid search, reranking, grounding, and honest evaluation of whether it is earning its complexity.
- ·Instrument the model layer — cost, latency, token spend, tool-call outcomes, failure taxonomy — and act on what it shows.
- ·Set the technical direction across two to four concurrent engagements and hold the standard in review.
- ·Work directly with client engineering teams: pairing, written architecture, and the handover that means they can run it without us.
- ·Carry production ownership for the systems you design, including on-call.
- ·Raise the bar for the rest of the team through code review and written decision records, not through status meetings.
Every line here is real.
- ·Senior-level Python engineering with production systems behind you, not notebooks
- ·Agentic framework experience is required — you have designed, shipped, and operated multi-step tool-using agents in production
- ·Solution architecture is required — you can own the design conversation with a client and defend the trade-offs in writing
- ·LLM systems in production: prompting as engineering, context strategy, tool schemas, structured output, and the failure modes of each
- ·Eval discipline — you can show how you measured an agent, not just that you built one
- ·Strong API and backend design; async Python, queues, and long-running workloads hold no surprises
- ·Comfort owning the whole path to production: containers, CI, deploys, observability
- ·Excellent written architecture — the design doc is the deliverable, not the meeting
- ·Directness with clients and the judgement to say no to the wrong requirement
- ·Based in the UAE and able to work a UAE-anchored schedule
- ·Depth with Claude and Claude Code specifically, or with MCP server design
- ·RAG at real scale, including retrieval evaluation rather than vibes
- ·Fine-tuning, distillation, or a considered view of when neither is warranted
- ·Data engineering background — pipelines, warehouses, and the messy upstream reality
- ·Regulated-industry delivery: fintech, government, healthcare
- ·Arabic in addition to English, including evaluation of Arabic-language output
- ·Open-source or public writing on agent architecture
One email, one reply within 5 days.
The button below fills in the subject line, so your application reaches the right principal.
Open email →Subject: Application - Lead AI Engineer · Python & Agentic Systems
- CVPDF preferred. One or two pages is plenty — we read all of it.
- WhyOne paragraph: why this role, and something you shipped end to end.
- LinksAnything public you are proud of. Optional, never penalised.