What should an AI engineer CV show?

Four things: models or LLM applications you shipped to real users, how you measured quality (evaluation results against a baseline), the MLOps work around them (deployment, monitoring, retraining), and the stack you used, tied to specific systems. Put production evidence first and name the scope honestly.

What's the difference between an AI engineer CV and a machine learning CV?

The dividing line is production, not terminology. An AI engineer CV needs to show you have shipped models to real users, handling inference latency at scale, distribution shift, evaluation pipelines, and fallback logic, not just that you experimented with models in a notebook. If your experience is on the experimentation side, say so rather than blurring the two.

What should an AI engineer CV include if I haven't shipped models to production yet?

Be honest about where you are and frame the adjacent engineering skills you bring: strong Python, API design, evaluation methodology, or MLOps tooling you have used outside a live production system. Precision about scope ('built a prototype, not production-deployed') reads as more credible than vague claims of production experience, and it survives a technical interview.

What metrics should I include on an AI engineer CV?

Cost and latency figures (p95 latency, requests per day, inference cost per call), evaluation results (F1, precision@k, or task-specific metrics versus a baseline), and reliability signals like drift monitoring or rollback logic. Use your real numbers. A specific figure, such as a measured p95 latency at a stated request volume, carries more weight than a general claim like 'optimized model performance'.

AI Engineer CV: What to Show and an Example

An AI engineer CV has to show four things: models or LLM applications you shipped to real users, how you measured whether they worked, the MLOps work that kept them running, and the stack you used, tied to specific systems. Lead with the production evidence. If a project was a prototype, say so; precise scope is more believable than a long list of frameworks.

Here is what that looks like on the page.

A worked example

The numbers below are illustrative. Replace them with your own measured results, and leave out any figure you cannot defend in an interview.

Summary

AI engineer with 5 years building and operating ML systems in production, specialising in LLM applications and RAG. Built and ran the retrieval and evaluation stack behind a customer support assistant used by thousands of customers. Focus on evaluation-driven development and inference cost.

Experience bullets

Designed and deployed a RAG support assistant on GPT-4o with a pgvector index of 80,000 articles; added re-ranking and citations, which cut average ticket resolution time by 35% against the previous flow.
Built an offline evaluation harness (200 labelled questions, groundedness and answer-correctness scoring) that gates every prompt and model change in CI; caught 3 regressions before release.
Fine-tuned Mistral 7B with QLoRA for contract clause classification, reaching 91% F1 versus 74% for the base model, served with vLLM on one A10 GPU at under 80ms p95 latency.

Each bullet names the system, the method, and a measured result.

The Production Gap

The main dividing line in AI engineering hiring is between people who have experimented with models and people who have shipped them. Experimenting means getting a notebook to produce interesting outputs. Shipping means handling distribution shift, inference cost, dataset and weight versioning, fallback logic for model failures, evaluation that catches regressions, and operating all of it reliably enough that users trust it.

If you have done the latter, make it unmistakable. If you have only done the former, be honest about it and frame the adjacent engineering skills you bring.

What Hiring Managers Scan For

The figures in the quoted phrasings below are placeholders to show the shape of a good line, not benchmarks.

1. Production deployment evidence. Not "trained a model on X dataset" but "deployed a classifier serving 50,000 requests per day at under 200ms p95." Not "fine-tuned GPT" but "fine-tuned a Mistral 7B model for entity extraction, deployed via vLLM on AWS, processing a legal document review pipeline daily." The specifics are the evidence.

2. Evaluation and reliability work. How did you know the model was performing well? Did you build evaluation harnesses, define task-specific metrics, run A/B tests against a baseline, or add regression tests to catch drift? Evaluation is the engineering work that separates experiments from production systems, and it is often missing from AI CVs.

3. MLOps and infrastructure. Experiment tracking (MLflow, Weights & Biases), model registries, automated retraining, deployment orchestration, feature stores, data versioning (DVC, LakeFS). If you built any of this, it belongs prominently on your CV.

4. LLM application architecture. For LLM product roles: RAG pipeline design, prompt engineering at scale, context window management, embedding model selection and evaluation, vector database choices (Pinecone, Weaviate, pgvector), and agent frameworks (LangChain, LlamaIndex, custom). Equally important is how you evaluated and controlled outputs, including guardrails and hallucination mitigation.

5. Cost and latency awareness. Inference is expensive. Token budgets, quantization, batching, caching embedding lookups and choosing the right model size for each task all signal production-grade thinking.

Key Skills to Highlight

Foundations and modelling:

Python (PyTorch, TensorFlow, JAX, scikit-learn, HuggingFace Transformers)
Core ML: supervised/unsupervised learning, transformer architecture, fine-tuning (LoRA, QLoRA, full fine-tune)
LLMs: GPT-4/o, Claude, Gemini, Mistral, Llama — API-based integration and self-hosted inference
Embedding models and semantic search; retrieval-augmented generation (RAG) architectures

MLOps and infrastructure:

Experiment tracking: MLflow, Weights & Biases, DVC
Model deployment: TorchServe, Triton Inference Server, vLLM, BentoML, Seldon
Orchestration: Kubeflow, Airflow, Prefect, Dagster
Vector databases: Pinecone, Weaviate, Qdrant, pgvector (PostgreSQL)
Cloud ML platforms: AWS SageMaker, Azure ML, Google Vertex AI

Evaluation and reliability:

Evaluation harness design, task-specific metric definition (BLEU, ROUGE, custom rubrics)
LLM judging frameworks, human feedback loops, RLHF basics
A/B testing for model versions, shadow deployment, canary rollouts
Monitoring: data drift detection, embedding drift, output distribution monitoring

LLM application stack:

LangChain, LlamaIndex, Semantic Kernel, custom agent frameworks
Prompt engineering patterns, few-shot design, chain-of-thought prompting
Guardrails: output validation, content filtering, structured output enforcement (Pydantic, Outlines)

Strong vs Weak Bullets

As above, the numbers in these examples are illustrative, not real results.

Weak: Deployed a model to production and monitored it. Strong: Added drift monitoring (population stability index on 12 input features, embedding drift on query text) to a ranking model, with alerts routed to on-call; detected a data-source change within a day and rolled back to the previous model version with no customer-facing incident.

Weak: Worked on machine learning pipelines and model training infrastructure. Strong: Built an end-to-end ML training pipeline on Kubeflow for a recommendation model retrained weekly on 200GB of behavioural data; implemented automated evaluation gates (precision@10, NDCG) with rollback logic — increased model currency from quarterly to weekly retraining cycles while reducing manual deployment effort from 6 hours to zero.

NextCV features — AI-tailored CVs, cover letters, and interview prep

Structuring Your AI Engineer CV

Professional summary. In 3 to 5 lines, state your specialisation (LLM applications, ML platform engineering, computer vision, NLP, recommendation systems), the kind of systems you have run in production, and the depth marker that sets you apart. The example near the top of this guide shows the shape.

Skills section. Organise by layer: modelling/frameworks, MLOps/infrastructure, LLM stack, cloud platforms. Do not sort alphabetically — sort by relevance to the roles you are targeting. Hiring managers scan this section for familiar terms and then validate them in the experience section.

Experience section. This is where AI CVs most often fail. Every role should have at least one bullet with a model or system outcome (accuracy, latency, throughput, cost). At least one bullet should describe the evaluation or reliability work, not just the training. If you contributed to infrastructure (pipelines, deployment, monitoring), make that explicit — it distinguishes you from pure researchers.

Projects and research section. AI engineering is a field where side projects, open-source contributions, papers, and Kaggle placements can add real weight. If you have any of these, a focused projects section is worth including. Link to GitHub, arxiv, or Kaggle as appropriate.

The Credentials Question

A 2026 AI engineer CV can have a wide range of formal credentials — a traditional CS/stats PhD, a bootcamp certificate, no credentials at all beyond demonstrated work. Credentials matter less than demonstrated work here, and interviewers tend to probe the technical evidence in depth.

Relevant credentials that carry genuine weight: an ML-focused masters or PhD from a recognised programme, Google Cloud Professional ML Engineer, AWS Certified Machine Learning Specialty, completion of fast.ai's Practical Deep Learning (with projects to show for it), or a strong Kaggle competition history. These are additive to demonstrated experience but do not substitute for it.

If you are mapping the wider field, start with the AI Product Manager CV guide, the guide to AWS on your CV, and the SQL for analysts CV guide.

Differentiating on Honesty

There is a temptation in the current AI market to overstate. Resist it. Technical interviewers at any serious AI engineering shop will probe claimed experience early in a screen. Being caught overstating your depth in PyTorch or LangChain is far more damaging than simply being honest about the level at which you have used each tool.

A more effective strategy: be precise about scope rather than vague about everything. "Used the OpenAI API to build a document classification prototype, not production-deployed" is more credible and more trustworthy than "worked with LLMs" — and it opens a conversation about what you learned and what you would do differently in a production context.

Tailoring for Different AI Engineering Roles

The AI engineering job market in 2026 contains several distinct role types that share a title but require different emphases:

LLM product engineer: Focus on RAG architecture, prompt engineering, agent frameworks, evaluation, and product iteration speed. Show user-facing outcomes.

ML platform engineer: Focus on training infrastructure, pipeline orchestration, feature stores, experiment tracking, and deployment automation. Show scale and reliability.

Applied ML scientist / research engineer: Focus on modelling depth, evaluation rigour, publication or applied research contributions, and domain expertise (NLP, vision, tabular, time-series).

AI infrastructure engineer: Focus on GPU cluster management, distributed training, inference serving at scale, model quantization, and cost optimisation.

Each of these profiles needs a different emphasis. Read the job description carefully: the infrastructure it describes and the team it sits within will tell you which profile the role belongs to. To see which of its keywords your CV is missing, paste the ad into the free job match tool.

Three steps to a tailored CV

NextCV reads job descriptions and maps your experience to the specific profile the role is targeting, so the right layer of your background gets the emphasis it deserves for each application.

Common Mistakes That Cost You Interviews

1. API calls presented as model development. Calling the OpenAI API is a valuable engineering skill but it is not the same as training or fine-tuning models. Be precise about where in the AI stack your experience lives.

2. No evaluation or reliability content. Many AI CVs list model training, and few describe how model quality was measured and maintained. Including evaluation methodology and reliability work instantly differentiates you.

3. Ignoring the software engineering layer. Many AI candidates underrepresent their software engineering skills because they are focused on the ML layer. But AI engineering roles need solid Python, API design, system architecture, and production operational skills. Do not leave these implicit.

4. Listing frameworks without context. "PyTorch, TensorFlow, LangChain, LlamaIndex, HuggingFace" — without any context about what you used them for, these list entries are nearly meaningless. Connect each tool to a specific system or outcome in your experience section.

Closing Thoughts

The AI engineers who get interviews are the ones who can show a model going from experiment to production, held to measured performance, and improved from what users actually did with it. Make that legible and specific on every page.

Want an ATS-readiness score first? Run your CV through the free CV checker →. It scores out of 100 across 8 categories and needs no job description.

Ready to put this into practice? Build your AI Engineer CV with NextCV →

For adjacent territory, see the Data Analyst CV guide and the per-posting tailoring walkthrough.