Hire an AI engineer

An AI engineer builds applications on existing language models (LLMs): an assistant that searches your documents, an agent that carries out steps in a workflow, or an LLM used as part of an analysis. They choose the model, build the connections to source systems and test whether the answers are correct.

If you’re looking for a GenAI engineer or LLM engineer, this is usually the same role. Our page on staff augmentation lists the AI engineer among the data science and AI roles.

An AI engineer builds on an existing language model and focuses mainly on sources, prompts, integrations and evaluation. A machine learning engineer trains custom models on your data and puts them into production. An MLOps engineer sets up the processes that keep all models running in production: registry, CI/CD and monitoring.

If you want predictions based on your own data, an ML engineer is the better fit. If you need an assistant to answer questions about documents, choose an AI engineer.

With retrieval-augmented generation, a search step first retrieves relevant passages from your documents. Those documents have already been split into chunks and stored as vectors using an embedding model. The language model formulates an answer based on those passages and cites the source, so you can check it.

At Fokker, we saw two pitfalls: a question with several parts may not retrieve all the relevant passages, and similar contracts for different clients can get mixed up. Splitting questions into parts and using clear metadata helps. Read more in our article on internal LLMs and RAG.

Use a test set of real questions and the answers you expect. For RAG, Microsoft distinguishes between two types of measurement: the search step (does the assistant retrieve the right passages?) and the answer. For the answer, you assess groundedness (does it contain anything that isn’t in the sources?), relevance and completeness.

Another language model can act as a judge and assign these scores. The AI engineer runs the test whenever the prompt, model or index changes, so you can see whether a change improves or breaks anything.

Start with the data you’ll put into it. If your organisation already uses Azure, Azure OpenAI is an obvious option: it comes with a standard data processing agreement, and you can have temporary storage of inputs disabled. Hosting an open model on your own infrastructure gives you the most control, but requires the most work and your own guardrails.

We discuss the trade-offs between privacy, hosting, costs and ease of use in our article how to choose the right LLM.

The biggest risk is prompt injection: text in a question, document or web page that gives the model different instructions. OWASP lists it as the top risk for LLM applications. The risk is greater with agents, which call systems themselves.

Among other measures, OWASP recommends giving the agent only the permissions it needs, marking external content separately, requiring human approval for high-impact actions and actively attacking the application in tests. Read more about agents under agentic AI.

People need to know they’re talking to a machine. According to the European Commission, the AI Act’s transparency rules apply from August 2026. If you use an assistant to make decisions about people, such as screening job applications, it may fall under the stricter rules for high-risk systems.

The AI engineer makes sure that answer sources, versions and tests are recorded. Your legal counsel or compliance team assesses whether an application falls under the rules. Read more under AI governance.

If you don’t yet have a defined use case, an engineer has nothing to build. Start with Data & AI strategy instead. If your documents are outdated or scattered, address the data foundation first. If you need a custom predictive model, a machine learning engineer is a better fit.

If you want an internal knowledge assistant but don’t have your own development team, take a look at Joost, our AI assistant for internal knowledge.

Read how we work with language models and maintenance data at an MRO company in AI in aircraft maintenance and MRO.