Hire AI Developers
This is the role most frequently mis-hired, because the title covers two different jobs. Almost everyone asking for an AI developer needs an engineer who builds systems around existing models, not a researcher who trains them.
Discuss your projectTwo different roles with one title
Being clear which you need will save you a great deal of time and money.
| AI engineer | ML researcher / scientist | |
|---|---|---|
| Does what | Builds products around existing models | Trains and researches models |
| Day-to-day | Retrieval, pipelines, evaluation, integration | Experiments, datasets, architectures |
| Core skill | Software engineering | Statistics and ML theory |
| Needed for | Nearly all commercial AI work | Novel capability, specialised domains |
| Cost | Comparable to a senior backend engineer | Substantially higher |
| Availability | Reasonable | Scarce |
If your project is a chatbot grounded in your documents, extraction from invoices, support triage, or any system using a commercial model through an API, you need an AI engineer. Hiring a researcher for that work is expensive and a poor use of their skills. We staff the first column, and would tell you plainly if your project genuinely needs the second.
What to interview for
The signal that matters most is evaluation discipline. Almost nothing else distinguishes a competent AI engineer from someone who has followed a tutorial.
- How they know whether it works. Ask what they would build first on a new AI feature. The answer you want is an evaluation set of real examples with correct answers, and a harness to measure against it. A developer who starts by writing prompts will adjust them for months without knowing whether anything improved.
- How they would measure accuracy on a specific task, and what accuracy the task actually requires. A good engineer asks what an error costs before asking how accurate it needs to be.
- Chunking strategy for retrieval, and why fixed-length chunks cut tables in half and separate clauses from their conditions. This is where retrieval quality is mostly won or lost.
- Hybrid retrieval. Whether they know semantic search is weak on exact identifiers — a part number, an error code — and that keyword search complements it.
- RAG versus fine-tuning, and that using fine-tuning to teach a model facts from your documents is a common and expensive mistake.
- What the product does when the model is wrong. The strongest signal of a product engineer rather than a demo builder: confidence thresholds, routing to a human, citations the user can check.
- Cost per request at volume, and whether they instrument it. AI unit economics kill more products than accuracy does.
- Underneath all of it, ordinary software engineering. An AI system is mostly a normal application with a model in it, and a developer who cannot write a reliable queue, handle retries or model data properly will produce something fragile whatever they know about prompting.
What they work on
Retrieval systems
Answering from your own documents with citations and permission filtering. The most common commercial AI project.
Document extraction pipelines
Structured fields out of invoices, contracts and forms, with validation and confidence routing rather than raw model output.
Classification and triage
Routing tickets, emails or documents against your own taxonomy, with a shadow period before anything goes live.
Evaluation infrastructure
The unglamorous work that makes everything else measurable. Frequently the highest-value thing an AI engineer does in their first month.
Voice and conversational systems
Integrating speech and language models with telephony and your own systems.
Model integration in existing products
Adding a capability to software you already run, which is mostly ordinary engineering with a model as one component.
Common questions
Related in Staff Augmentation
Tell us what you need
Describe the problem and we will come back with scope, timeline, and a rate. No obligation to proceed.
Get in touch