AI Engineer

About the role

You will join a small senior crew embedded inside a client organisation, usually a bank, telco, health system or energy operator. You will work in their repository, on their CI, in their on-call rotation. The systems are consequential and the constraints are real: regulatory scrutiny, legacy integration, and a business that cannot tolerate a bad release.

Most of this work is engineering, not modelling: retrieval, evaluation, latency, cost and failure modes. You will be expected to be honest about what these systems cannot do, particularly where a wrong answer has consequences.

The team you join

You will work with an embedded crew and directly with the business owners of whatever decision the system touches.

What you will do

• Build and ship LLM-backed features against real business problems.
• Design evaluation so quality is measured rather than asserted.
• Build the data and retrieval pipelines these systems depend on.
• Own latency, cost and reliability in production, not just accuracy.
• Advise clients candidly on where machine learning is and is not the answer.

Minimum qualifications

• Five or more years building production software, with recent applied ML work.
• Practical experience with LLM applications beyond prototypes.
• Strong Python or TypeScript, and solid data engineering fundamentals.
• Rigour about evaluation, and scepticism about demos.
• The judgement to refuse a use case that should not be automated.

Preferred qualifications

• Experience with regulated decisioning, such as credit or claims.
• Familiarity with vector search and retrieval-augmented generation at scale.
• Exposure to model governance and explainability requirements.

Your first three months

1. Establish how quality will be measured before building anything.
2. Ship a narrow, genuinely useful capability within six weeks.
3. Instrument cost, latency and failure modes from the start.
4. By month three, be able to say precisely where the system should not be trusted.

Compensation and benefits

Rates are agreed per engagement and depend on scope and seniority. We discuss numbers openly on the first call, before you spend time on a practical.

• Rates are agreed per engagement and discussed openly on the first call, before you spend time on a practical.
• Remote across Africa, with on-site weeks where the engagement calls for them.
• Work embedded inside regulated institutions, on systems with real consequences, alongside senior engineers who review your work and expect you to review theirs.
• Small crews and direct access to partners, with no layers between you and the decisions.
• Long relationships. We would rather work with the same capable people across engagements than hire for a single project.
• Straight answers. We tell you what we can and cannot offer before you commit, and we do not promise what we cannot deliver.

How we interview

The whole process is normally two to three weeks, and we work around your current job.

1. Application review. A partner reads every application. You hear back either way, usually within two working days.
2. Intro call, 45 minutes. Your background, what you want next, and honest answers about the work, the rate and the constraints.
3. Practical session, 4 to 6 hours. A realistic problem close to what the role actually involves, scheduled around you and sized to respect your time.
4. Practical review, 60 minutes. We walk through what you built, why, and what you would change with more time.
5. Partner conversation, 45 minutes. Scope, expectations, terms, and your questions.
6. Decision, normally within two working days of the last conversation.

How to apply

Send an application through the form. A partner reads every one, and you will hear back either way. If you are unsure whether you qualify, apply anyway and say what you are unsure about.

We will tell you what we can and cannot offer before you commit to anything.

Apply for this role →