Prompt engineer CVs have a proof problem. Every candidate who has used ChatGPT, Claude, or Gemini can claim to have "experience with large language models." The claim is almost universally true and almost universally undifferentiated. The candidates who stand out are those who can demonstrate systematic, measurable prompt engineering — not iteration by intuition, but optimisation by evaluation: A/B testing prompts against a labeled dataset, measuring retrieval precision in a RAG pipeline against a ground truth benchmark, red-teaming a system prompt and documenting the vulnerabilities found, or using DSPy to treat prompts as learnable parameters optimised against a metric. The prompt engineering market in 2026 has also separated from the 2023 moment when companies were hiring anyone who could string together a few-shot prompt in ChatGPT. The current market wants AI engineers who build production systems: RAG pipelines with thoughtful chunking strategies and re-ranking, agentic architectures with tool use loops and error recovery, and evaluation infrastructure that treats prompt quality as measurable, not subjective. This guide covers the evaluation vocabulary, RAG architecture specifics, and model API expertise that makes a prompt engineer CV credible in 2026.

What Prompt Engineer Job Descriptions Require in 2026

Prompt engineer JDs in 2026 span a range from highly technical AI engineering to applied AI product roles:

Production AI systems engineering: Building and maintaining LLM-powered features and pipelines. JDs specify: Python proficiency (the primary language for all LLM tooling), REST API experience with LLM APIs (Anthropic's Messages API, OpenAI's Chat Completions API, Google's Gemini API), LLM orchestration frameworks (LangChain, LlamaIndex, LangGraph for agentic workflows), RAG implementation experience (vector databases — Pinecone, Weaviate, Chroma, PGVector — embedding model selection, chunk strategy, retrieval evaluation), and systematic evaluation frameworks (LangSmith, Braintrust, PromptFoo, or RAGAS for measuring prompt and retrieval quality). Senior JDs add agentic system design (multi-agent orchestration with AutoGen, CrewAI, or custom orchestrators), fine-tuning experience (LoRA/QLoRA parameter-efficient fine-tuning via Hugging Face PEFT), and the ability to write and maintain evaluation datasets.

Applied AI and model customisation: Optimising LLM behaviour for specific domains. JDs specify: few-shot example curation, system prompt design and version control, output format specification (XML structured prompts for Claude, JSON schema function calling for OpenAI), temperature and sampling parameter optimisation, red-teaming and safety evaluation, hallucination detection and mitigation strategies, and guardrail implementation (NeMo Guardrails, Llama Guard, custom classifiers). These roles sit between pure engineering and applied research.

AI product and developer relations: Technical writing and advocacy for AI products. JDs specify: the ability to write code samples that work (verified, executable code in Python, JavaScript, and cURL), documentation of LLM API capabilities and limitations, tutorial creation for developer audiences, and community engagement around AI tools. These roles require genuine API implementation depth — the writer must be able to run the code.

Cross-type requirements: Python, REST API consumption, Jupyter notebooks for experimentation, Git for versioning prompts and evaluation datasets as code, and a demonstrated portfolio — GitHub repos with real prompt engineering work, published blog posts on LLM evaluation, or open-source contributions to LLM tooling.

Prompt engineer salaries in 2026: £55K–£90K UK; £80K–£130K for senior AI engineers with production RAG experience. US: $110K–$170K; $150K–$230K for senior AI engineers at foundation model companies.

ATS Keywords for a Prompt Engineer Resume

Prompt engineer ATS filtering uses model names, framework names, and evaluation methodology vocabulary. "Generative AI experience" as a standalone phrase is the weakest possible signal.

Essential ATS terms for a prompt engineer resume:

  • Title variants: Prompt Engineer, AI Engineer, LLM Engineer, Generative AI Engineer, AI Product Engineer, Applied AI Engineer, Machine Learning Engineer (Applied), AI Solutions Architect, Developer Relations Engineer AI
  • Core disciplines: prompt engineering, large language model, LLM, generative AI, foundation model, retrieval-augmented generation, RAG, AI agent, agentic AI
  • Foundation models: Claude, Claude 3.5, Claude 4, GPT-4, GPT-4o, o1, o3, Gemini, Gemini Pro, Gemini Flash, Llama, Llama 3, Mistral, Mistral Large, Command R, Cohere, Qwen, Phi-3
  • AI companies and APIs: Anthropic, OpenAI, Google AI, Mistral AI, Cohere, Hugging Face, Ollama, vLLM
  • Orchestration frameworks: LangChain, LlamaIndex, LangGraph, DSPy, Semantic Kernel, AutoGen, CrewAI, Swarm
  • RAG components: vector database, embeddings, Pinecone, Weaviate, Chroma, PGVector, Milvus, Qdrant, chunk size, overlap, re-ranking, cross-encoder, BM25, hybrid retrieval
  • Prompt techniques: chain-of-thought, CoT, few-shot, zero-shot, function calling, tool use, structured output, JSON mode, XML prompt, system prompt, meta-prompting, ReAct
  • Evaluation tools: PromptFoo, Braintrust, LangSmith, RAGAS, evaluation dataset, eval, red-teaming, hallucination detection, accuracy, retrieval precision, F1
  • Safety and alignment: red-teaming, jailbreak, prompt injection, adversarial prompting, NeMo Guardrails, Llama Guard, constitutional AI, RLHF, DPO, alignment
  • Fine-tuning: fine-tuning, LoRA, QLoRA, PEFT, Hugging Face, parameter-efficient, SFT, supervised fine-tuning
  • Tools: Python, Jupyter, REST API, Git, GitHub, JSON, YAML, streaming, context window, token
  • Long-tail phrases: prompt engineer resume, prompt engineering resume, AI engineer resume, LLM engineer resume, how to write a prompt engineer resume, generative AI engineer resume, prompt engineer cv 2026

Placement: Model names (Claude, GPT-4o, Llama 3) in the Skills section — these are the primary ATS filter terms. Evaluation methodology (eval dataset, PromptFoo, retrieval precision) in experience bullets — this is the differentiation signal. RAG component specifics (vector database name, embedding model, chunk strategy) wherever RAG was implemented — the generic "LangChain and Pinecone" is weak; the specific "Cohere Embed v3 with 512-token chunks, Weaviate hybrid retrieval, cross-encoder re-ranking" is strong.

Prompt Engineer CV Structure and Bullets That Show Systematic Evaluation Depth

Section order:

  1. Headline — "AI Engineer | Prompt Engineering · RAG · LangChain · Claude/GPT-4o · PromptFoo · Python"
  2. Skills — Foundation Models & APIs / Orchestration Frameworks / RAG Components / Evaluation Tools / Fine-tuning / Languages & Tools
  3. Experience — 4–5 bullets per role; model and API used, architecture built, evaluation metric and score, and measurable product outcome per bullet
  4. Projects / Portfolio — GitHub repos with prompt libraries, eval datasets, or open-source LLM tooling; blog posts on LLM evaluation or RAG architecture; link to GitHub profile

Two pages for 4+ years; one page for roles focused entirely on prompt engineering without broader engineering depth. GitHub profile URL in the header — reviewed before every AI engineering interview.

Three elements make a prompt engineer bullet convincing: the model and API specifics, the system architecture built (RAG, agent, eval pipeline), and a measured outcome (retrieval precision, accuracy, CSAT improvement, latency). Three examples:

  • Built a production RAG pipeline for a B2B legal research SaaS — Claude claude-sonnet-4-6 with 200K context window for long-document synthesis, Cohere Embed v3 English for document embedding with 512-token chunks and 64-token overlap, Weaviate vector database with hybrid BM25 + dense retrieval and cross-encoder re-ranking; evaluated on a 340-question labeled dataset: 84% retrieval precision and 79% answer accuracy versus 61% baseline with naive nearest-neighbour retrieval; pipeline serves 2,400 monthly active researchers
  • Designed and maintained the system prompt suite for a customer support AI agent — 12 category-specific prompts (billing, technical, returns, escalation, compliance) with XML-structured output templates for reliable JSON extraction, and curated few-shot examples per category; A/B tested prompt versions using PromptFoo against a 500-interaction evaluation dataset; customer satisfaction for AI-handled tickets from 3.2 to 4.1 on a 5-point scale, and first-contact resolution from 42% to 67% over 8 weeks
  • Red-teamed an enterprise HR software company's customer-facing LLM deployment — tested 340 adversarial prompt patterns covering jailbreaks, prompt injection, data extraction attempts, and policy violations; identified 14 system prompt vulnerabilities with documented proof-of-concept; implemented NeMo Guardrails for policy enforcement and re-tested all 14 findings; policy violation rate in production monitoring fell from 8.3% to 0.4% over 30 days

Prompt engineer interviews include practical exercises (here is a broken prompt — fix it and explain your reasoning), system design questions (design a RAG pipeline for this use case), and evaluation discussions (how would you measure whether this prompt change is actually better?). Your CV's evaluation methodology and RAG architecture evidence determine whether the interview explores your AI systems thinking or tests your basic prompting knowledge.

Three Prompt Engineer CV Mistakes That Signal Hobbyist AI, Not Production AI

No evaluation framework or accuracy metrics. "Designed prompts for a customer support chatbot" and "designed prompts for a customer support chatbot that increased first-contact resolution from 42% to 67%, measured via A/B testing against a 500-interaction evaluation dataset using PromptFoo" describe the same work at completely different levels of engineering rigour. Anyone who has used an LLM can claim to have designed prompts. The candidates who distinguish themselves are those who treated prompt design as an engineering discipline with measurement: a labeled evaluation dataset, a defined success metric (accuracy, F1, win rate, retrieval precision, CSAT), and a tool for running evals systematically (PromptFoo, Braintrust, LangSmith, RAGAS, or a custom eval harness in Python). If you have done this work, the metrics belong in your CV bullets. If you haven't, building a small eval dataset and running systematic prompt comparisons is a weekend project that will immediately separate your CV from the majority of prompt engineer applicants.

RAG pipeline described without architecture specifics. "Built a RAG chatbot using LangChain and Pinecone" appears on thousands of AI engineer CVs in 2026 — it has become the equivalent of "built a web app using React and PostgreSQL" in 2018: technically true, but providing no differentiation. The architecture decisions in a production RAG system are where the engineering judgment lives: chunk size and overlap strategy (how did you decide on 512 tokens with 64-token overlap, and what did you evaluate against?), embedding model selection (ada-002 vs. bge-large-en-v1.5 vs. Cohere Embed v3 — why?), retrieval strategy (dense vector only, BM25 only, or hybrid — and how did you tune the weighting?), and re-ranking (cross-encoder, reciprocal rank fusion, or LLM-as-a-judge). Include one specific architecture decision per RAG bullet — the fact that you made a deliberate choice and can explain it is the engineering signal the generic "LangChain + Pinecone" doesn't carry.

No model-specific API knowledge named. "Experience working with large language models" describes someone who has opened a chat interface. "Implemented multi-turn conversation management with Claude's Messages API, using XML-structured prompts for reliable JSON extraction and context window management for 200K token research documents" describes a developer with hands-on implementation knowledge of a specific API's behaviour. Name the model, the API feature, and the implementation challenge it addressed: function calling with JSON schema constraints in the OpenAI API, tool use with XML tags in the Anthropic API, system instruction grounding in the Gemini API, or streaming responses with partial JSON parsing. These specifics confirm that the candidate has written code against the API, not just read the documentation.

If you are applying to prompt engineer or AI engineer roles and want your CV rebuilt around the specific model APIs, RAG architecture, and evaluation requirements in a target job description, Resumegpt generates your AI engineer CV from your work history in under 60 seconds — model API expertise named, RAG architecture evidenced, evaluation methodology signalled, ATS-optimised, and exported as a PDF ready to submit.