01Owns
Retrieval quality
Chooses chunking, embeddings, hybrid keyword and vector search and re-ranking, and measures whether the right passages actually reach the model.
Hire · AI engineering
Hire Generative AI Developers
Senior generative AI developers who join your team to own features built on large language models: retrieval over your documents, tool calling into your systems, and the evaluation sets that tell you whether a prompt change helped. Every candidate has fixed a failing retrieval pipeline and written graded test cases in a paid trial.
Trial graded on an evaluation set, not a demo
Senior and lead engineers only
Provider-neutral: Claude, GPT, Gemini or open models
On your team
What a generative AI developer owns on your team.
02Owns
Prompts and tool definitions
Writes system prompts, structured output schemas and tool descriptions as versioned code, with tests that catch regressions when wording changes.
03Owns
Evaluation sets
Builds graded examples from real user queries, combines exact checks with model-graded and human review, and runs them on every change.
04Owns
Guardrails and safety
Handles prompt injection from retrieved content, limits what tools can do, filters sensitive data and decides when the system should refuse or escalate.
05Owns
Cost and latency
Tracks tokens per request, uses prompt caching, streaming and smaller models for simple steps, and keeps response times inside an agreed budget.
Demos are easy; evaluation is the job
Almost anyone can build an impressive chatbot demo in an afternoon. The hard part starts when real users ask questions the demo never saw, and every prompt tweak fixes one case while breaking two others. The engineer you want is the one who builds a test set before changing anything.
In interviews, ask how a candidate decided their last LLM feature was ready to launch. Listen for numbers from a graded set, categories of failure and how they handled disagreement between automated judges and human reviewers.
Vetting
Skills we test before a generative AI developer joins.
01Tested
RAG engineering
Index design, metadata filtering, query rewriting and diagnosing whether a bad answer came from retrieval or from generation.
02Tested
Tool use and agents
Designs narrow tools with clear schemas, bounds loops, handles tool errors gracefully and knows when a fixed workflow beats an autonomous agent.
03Tested
Evaluation practice
Writes rubrics, calibrates LLM judges against human labels and tracks pass rates per category rather than a single average.
04Tested
Provider and model choice
Compares hosted APIs and open-weight models on quality, context length, price, data residency and rate limits for your workload.
05Tested
Security for LLM apps
Familiar with the OWASP Top 10 for LLM applications, including indirect injection, excessive agency and leakage of system prompts.
06Tested
Production plumbing
Streaming responses, retries with backoff, idempotent tool calls, tracing of each model call and redaction in logs.
Work sample
How the generative AI developer work sample runs.
- 01Step 1
Diagnose weak answers
The candidate gets a document assistant that answers confidently but wrongly, traces the failure to chunking and retrieval, and shows the fix with before-and-after scores.
- 02Step 2
Write the evaluation first
From thirty real-looking queries they build a graded set with expected facts and failure categories, then use it to judge two prompt variants. Paid at our standard rate.
- 03Step 3
Add a safe tool
They give the assistant one tool that reads order status, with input validation, permission checks and a refusal path when the user asks for someone else’s data.
- 04Step 4
Explain the bill
A short call estimating per-conversation cost and latency for the design, and naming two ways to reduce each without hurting quality.
Levels
Generative AI Developers: what changes at each level.
01Level 1
Senior
Ships and maintains LLM features within an existing stack, owns their evaluation sets and reviews prompt and retrieval changes from peers.
02Level 2
Lead
Sets the evaluation framework, prompt and tool conventions, provider strategy and the rules for what reaches production.
03Level 3
AI platform lead
Builds shared infrastructure for multiple teams: model gateways, tracing, evaluation pipelines, cost controls and policy enforcement.
Senior, lead or platform lead
Choose a senior engineer when one product team is adding LLM features to an existing application. Bring in a lead when several features are in flight and nobody agrees how prompts are versioned, evaluated or approved. An AI platform lead is warranted once multiple teams call models directly and you need shared routing, tracing, budgets and policy.
Other options
When to hire someone other than a generative AI developer.
01Consider
You need to train predictive models
Forecasting, scoring and classification on tabular data is classic machine learning; a data scientist fits better.
Open register02Consider
The AI part is a single API call
If the feature is one summarisation call inside a larger product, a strong full-stack developer can own it.
Open register03Consider
Your assistant lives on Azure
If you are committed to Azure OpenAI with Entra ID and Azure AI Search, an Azure developer may integrate faster.
Open registerSeparate retrieval failures from model failures
When a retrieval-augmented answer is wrong, the cause is usually upstream: the right passage was never retrieved, was split mid-sentence or was outranked by a near duplicate. Strong developers log what was retrieved for every answer and measure recall on that step separately, rather than rewriting the prompt in hope.
What to settle in the opening weeks
Agree the provider terms, data retention settings and which data classes may enter a prompt before the first commit. These decisions are slow to revisit later.
Then ask for a baseline evaluation of whatever already exists, even a prototype. A measured starting point turns every later discussion from opinion into comparison.
FAQ
Questions to resolve before the build.
01Which model providers do your generative AI developers work with?
Most have shipped on at least two of Anthropic, OpenAI, Google and open-weight models served through vLLM or a managed host. We vet for the reasoning behind the choice rather than loyalty to one vendor.
02How do we know an LLM feature is getting better?
Through an evaluation set built from real queries and run on every change. Your developer owns that set, reports pass rates by category and adds a new case whenever a user finds a failure.
03Can they work with our confidential documents?
Yes, within your environment and access controls. They will ask about data residency, retention settings at the provider and which content must never reach a prompt before building anything.
Works alongside
Roles a generative AI developer works best alongside.
01Pair with
Python developers
For ingestion jobs that keep the document index fresh and the APIs the assistant calls through its tools.
Open register02Pair with
UI/UX designers
To design streaming answers, citations, uncertainty and correction flows so users trust the output appropriately.
Open register03Pair with
QA engineers
To add exploratory and adversarial testing on top of automated evaluations before each release.
Open registerProduct delivery
Need the whole product built instead?
Primary sources
References behind this page
Dated official documentation, standards, and research that support the factual claims on this page.
- 01Anthropic documentation: Create strong empirical evaluations
Guidance on building test cases and grading methods for LLM applications.
- 02OWASP Top 10 for Large Language Model Applications
Security risks specific to LLM apps, including prompt injection.
How to read this page
Scope, cost and timeline figures are planning guidance until confirmed in a written proposal. Diagrams and screens are illustrations, not client deliverables, and source-code rights and handover follow the signed agreement. See the delivery process and the full disclaimer.
Next decision
Turn the brief into an accepted product scope.
Define outcomes, constraints, evidence, rights and handover before delivery begins.
Commercial rights, repositories, environments, documentation, acceptance and handover remain contract-defined.