AI Engineer
You’d work on what surrounds the AI model, not the model itself: the company context that reaches it before anyone types, routing each job to the right model, and tests honest enough to show that a cheaper model already does the job.
Most of the week goes into making AI useful inside one specific company: bringing their real documents and methods in, writing down what a good result looks like for a job they do every day, and then proving the results improved. Some of it is retrieval and prompt design. A lot of it is measurement, because without a clear quality bar, choosing a model is just guessing with confidence.
- Build company, role and personal setups, and the rules that stop a personal tweak from overriding a company rule.
- Define a quality bar for each job, then build the tests that check whether a run met it.
- Route work across model tiers so routine steps stop running on the most expensive models, with costs tracked by app and by person.
- Ship the named jobs people actually use, in chat and inside a client’s own product.
- Write down what you learned so the next person starts where you left off. The project notes are part of the job, not paperwork afterward.
- Strong Python, and enough TypeScript to be useful where the AI meets the interface.
- Real experience putting an LLM system into production: retrieval, tool use, structured output, handling failures. Papers are welcome. Something that ran is better.
- An instinct for evaluation. If you’ve ever argued that a demo proved nothing, you’ll fit in.
- Plain writing. A finding an owner can’t act on isn’t finished.
- Comfort writing “I couldn’t verify this” to a client.
Not required: a degree in the field, loyalty to any agent framework, or benchmark trivia.
- How often the first result on the jobs you built gets approved. Measured, not claimed.
- The share of work a cheaper model can be trusted with. It should keep rising.
- Whether your setup works for the next client too. That’s worth more than one that impressed this one.
- Show us: Use the form below to send one thing you’ve built and are proud of. We read everything that comes in, and if there is a fit we’ll come back within a week.
- Work through something together: An hour or two with the people you’d work beside, on a real problem from our work.
- Decide: A conversation with the two people you’d work beside, then a clear yes or no, with the reason.