LLM Engineer
You build applications on top of large language models — retrieval, agents, evaluation and serving.
Degree usually expected Moderate for a working software engineer. The hard part is evaluation and reliability, not calling an API.
Can I actually do this?
Not a first tech job. For someone who already ships software it is a shorter step than the research-facing roles, because much of the work is ordinary backend engineering applied to a model you did not train — that is our reasoning from the skill overlap, not a hiring-volume source we have verified.
Who it suits. Software engineers who are comfortable with systems whose output is probabilistic and who like building guardrails around that.
Runway. Months rather than years if you already write production software; this is an applied layer, not a research one.
Coming from backend, data or ML engineering? That is the assumed starting point — this role adds a specialism rather than replacing your skills. Data Engineer (AI/ML) Machine Learning Engineer
Also advertised as
The route
Four stations, in order. Each one is a thing you finish before the next matters.
-
Station one
Learn it free
Only the best few, deliberately. Every one of these is free to use — the pill on each card says exactly what is and isn't free.
Hugging Face LLM Course
Free to learn · no certificate
Transformers, fine-tuning, datasets and tokenizers, hands-on. Hugging Face states it is "completely free and without ads". IMPORTANT: their FAQ says plainly "Currently we do not have any certification for this course" — their separate Agents Course does certify, this one does not. Widely-copied listings get this wrong.
Verified 2026-07-26
Anthropic's educational courses (open source)
Free to learn · no certificate
Anthropic's own course repository covering API fundamentals, prompt engineering and evaluation. Free and openly licensed on GitHub; no certificate.
Verified 2026-07-26
vLLM quickstart — serving an LLM at scale
Free to learn · no certificate
Serving is the part that separates a demo from a product. NOTE THE URL: docs.vllm.ai/en/latest/ serves developer-PREVIEW docs and says so on the page — this is the stable release path.
Verified 2026-07-26
-
Station two
Attest strategically
There is no established, widely-recognised LLM-engineering certificate yet, and we are not going to invent one for you. Google's ML Engineer credential is adjacent but assumes 3+ years of industry experience. In this role a working application with evaluation results is the credential.
Google Cloud Professional Machine Learning Engineer
Google Cloud · Professional ML Engineer
Free to learn · paid certificate Recognized
The honest cost Cost $200
Plus tax where applicable. Printed on Google's own exam page.Validity 3 years. Renewal Renew within the eligibility period before expiry; Google publishes shorter renewal exams for its certifications. Assessment Proctored exam — two hours, 50–60 multiple-choice and multiple-select questions, online- or onsite-proctored Proctored Yes Verify via unknown Cost per active year $67/yr200 ÷ 3 Google's role-mapped ML credential. Recommended experience is 3+ years in industry including 1+ year on Google Cloud — the credential assumes you are already working.
Read this before you buyPrerequisites are formally None, but Google's own recommended experience is 3+ years in industry. Passing this without that background is possible and unusual; it is not an entry-level route into AI work.
Verified 2026-07-26 · Official page
-
Station three
Prove it
A certificate says you passed a test. These say you can do the job.
A retrieval application with measured quality
Build a RAG application and evaluate it — retrieval precision, answer faithfulness, failure cases. The evaluation is the artifact; anyone can wire up a vector store.
An agent with real guardrails
Build an agent that calls tools, and document what happens when a tool fails, a loop runs long, or the model returns nonsense. Handling those is the engineering.
A served model with cost and latency numbers
Serve a model behind an API with measured p95 latency and cost per thousand requests. Publish the numbers and what you changed to improve them.
-
Station four
Get hired
Search these exact titles
Who hires for this. AI-native startups, product companies adding AI features, consultancies, and internal platform teams at larger firms.
Evaluation results are the differentiator. Plenty of people can demo a chatbot; far fewer can show what its failure rate is and how they measured it — that is our reasoning about what is checkable, not a hiring statistic we have verified.
On salaryWe don't publish salary estimates. Numbers copied between blogs drift from reality, and a wrong number costs you real negotiating power. When we have a verified public source, it goes here with its date.
Where this route continues
- AI Product Engineer — toward owning what ships
- Machine Learning Engineer — deeper into modelling
- AI Security Engineer — toward evaluation and safety
- Machine Learning Engineer — sideways move
- Data Engineer (AI/ML) — sideways move
Continues into ML engineering and AI platform work.
This page last verified 2026-07-26 · How we verify
All three resources fetched and READ 2026-07-26. Hugging Face's no-certification statement and its "completely free" claim are quoted from the course page itself; the Agents-course certificate was separately confirmed live. The vLLM preview-vs-stable split was found by reading the banner on the /latest/ page.