AI, data & engineering
NLP Engineer
NLP Engineer is an engineer who trains and fine-tunes language models for a product task when an off-the-shelf LLM falls short on accuracy, speed or privacy.
Also known as: Natural Language Processing Engineer, NLP Scientist, Applied NLP Engineer, NLP Research Engineer, Computational Linguist.
What does an NLP Engineer do?
An NLP Engineer is an engineer who builds text systems at the model level: they train and fine-tune language models, collect text data and measure quality on a specific task. Many text tasks are now solved by a request to an off-the-shelf LLM, and an NLP Engineer is needed where that is not enough. Typical tasks:
- fine-tune language models for a domain and language, and distil large models into small, fast ones;
- build classification, entity extraction, search and ranking of texts;
- collect and clean text corpora and organise labelling;
- work with languages where off-the-shelf models are weak and with narrow terminology: medicine, law, finance;
- compare their own model with an off-the-shelf LLM on quality, latency and price and defend the choice with numbers.
When does a company need an NLP Engineer?
A company needs an NLP Engineer when text is the core of the product and an off-the-shelf LLM via API does not fit on accuracy, price, latency or data requirements. Typical situations: millions of documents a day, where calling a large model on each one is too expensive; a closed environment where data cannot go to an external provider; a language or domain where off-the-shelf models make mistakes. The role will not work if prompts and RAG over an existing model solve the task and nothing needs training: then you hire an AI / LLM Engineer.
NLP Engineer vs AI / LLM Engineer: what's the difference?
An NLP Engineer differs from an AI / LLM Engineer in working at the level of the model rather than the product around it: they train, fine-tune and compress language models, while an AI / LLM Engineer builds a product on existing ones through prompts, RAG and agents. What sets them apart from an ML Engineer is the specialisation in text: tokenisation, the specifics of languages, evaluating quality in natural language.
NLP Engineer vs related roles: comparison
| NLP Engineer | AI / LLM Engineer | ML Engineer | |
|---|---|---|---|
| Responsible for | Language models for the task | A product on existing LLMs | Models on any data |
| Main result | A model more accurate, faster or cheaper than an off-the-shelf one | A working LLM feature | A model in production |
| Main tool | Training, fine-tuning, distillation | Prompts, RAG, agents, evals | Training and serving pipelines |
Why is it hard to hire an NLP Engineer?
NLP Engineers are hard to hire because within a few years the market split into people who worked with text before LLMs and people who arrived in the LLM era, while a company usually needs someone who understands both.
- Many strong specialists have changed their title to AI Engineer or LLM Engineer, and searching by the old name does not find them.
- The reverse problem: many AI Engineers have only worked with APIs and never trained a model, and a CV does not show the difference.
- Experience with a specific language or domain, such as medical or legal texts, is rare.
What to decide before hiring an NLP Engineer?
- Is an off-the-shelf LLM enough? If it is, you need an AI / LLM Engineer, and an NLP Engineer will be left without training work and leave.
- Which language and which domain? Experience with English corporate text does not automatically carry over to another language or to medical records.
- What matters most: price, latency or data privacy? The answer decides whether you need experience in distillation, optimisation or deploying models in a closed environment.
Where to find an NLP Engineer?
NLP Engineers most often come from search companies and marketplaces (search, ranking, moderation), legal tech and medtech that process documents, and machine translation, voice assistant and chatbot companies. A strong signal is a model the person trained or distilled for a product task themselves, with an honest comparison against an off-the-shelf LLM on quality and price. NEWHR finds such candidates by the results of their work, not by applications to a job post.
How to assess an NLP Engineer in an interview?
NLP Engineers are assessed with a review of a text model the candidate took to production and a technical case in which they choose an approach for the company's task: a prompt to an existing LLM, fine-tuning or their own model. On the company side, an ML Lead or a lead ML engineer takes part, and for a product task, a Product Manager.
Strong signals:
- first checks whether an off-the-shelf LLM solves the task, and only then proposes training;
- explains how they collected and checked the data, not only which architecture they chose;
- compares options on quality, latency and cost with concrete numbers.
Red flags:
- proposes training their own model where a prompt would do;
- has not worked with LLMs and treats them as a fad;
- cannot explain how they measured quality on text beyond overall accuracy.
How long does it take to hire an NLP Engineer?
We present the first candidates for the NLP Engineer role within 5–7 days of the brief.
Related roles
- AI / LLM Engineer — the first hire for most text features; an NLP Engineer comes in when prompting an existing model is no longer enough.
- ML Engineer — fits when text is a small part of the product and the main models run on behavioural and tabular data.
- Evals Engineer — designs the criteria for judging a text model's answers, and the NLP Engineer improves the model against them.
- Voice AI Engineer — owns audio and latency in a spoken conversation; an NLP Engineer's text models can run inside that chain.