Refery for hiring teams

Hire an AI engineer

Find an engineer who can make AI useful in a production workflow and explain how they know it works.

For founders, CTOs, and heads of talent hiring engineers to build and operate AI product capabilities. Begin with the user workflow, the acceptable failure modes, and the evidence required to release.

The shortlist, with the reasoning
Ethan ChoCo-founder intro · work reviewed
A+
Ines CarvalhoSpecialist search · screened
A−
Lena BrandtScout intro · screened
A

An illustration of how a Refery search is presented. The names and grades are examples, not real candidates.

Our clients are backed by investors including:

  • Y Combinator
  • Sequoia
  • a16z
  • General Catalyst
  • Lightspeed
  • Founders Fund
  • Greylock
  • Insight Partners

Make the role clear

Decide what the job is before the search starts.

Applied AI product work, model research, inference infrastructure, and customer deployment are separate hiring problems. A single title does not make the experience interchangeable. Choose the work that will occupy the first months.

Put these decisions in the brief

  • The user workflow and current non-AI baseline
  • Available data, access boundaries, and quality constraints
  • Evaluation criteria, latency needs, and operating budget
  • Ownership of deployment, monitoring, and human escalation
The brief
Must own
The user workflow and current non-AI baseline
Stage
VC-backed, Seed to Series B
Still open
Location, band, and the first milestone
Scope and fee agreed before the search15%

What to look for in candidate evidence

Three things a shortlist should show.

Evaluation

A task-specific evaluation set and a failure that changed the design.

Software delivery

A production integration, its tests, and the candidate’s personal contribution.

Operational judgment

How quality, latency, and cost were measured and what happened when the system failed.

A useful interview exercise. Give the candidate a small, fictional workflow and a few failure examples. Ask them to propose a baseline, an evaluation method, a release decision, and a fallback. Discuss which parts need ordinary software engineering and which benefit from a model.

Start a search with Refery

One brief. Two networks. A reasoned shortlist.

Refery is a startup recruiting network for VC-backed Seed to Series B companies hiring Engineering, AI, and GTM talent. It combines specialist independent recruiters with referrals from founders, operators, and investors.

Refery charges 15% to 25% of first-year base salary on a successful hire. No retainer, no exclusivity, and a 90-day guarantee under the client agreement. Scope and fee are agreed upfront. See current pricing.

Send your hiring brief →
  1. Share the brief

    Send the role, location, compensation, and the outcome the hire needs to own.

  2. Confirm fit and scope

    Discuss the search, candidate evidence, interview process, and commercial terms with Refery.

  3. Keep the decision grounded

    Use a consistent scorecard, capture interview evidence, and give specific feedback as the search progresses.

Before you set the budget

Compare like with like.

Compare roles with the same responsibilities and working arrangement. Our salary guides show dated public posting examples, with base salary, equity, and variable pay kept separate.

A question hiring teams ask

Should we require experience with a particular model provider?

Require it when the existing system makes that experience necessary. Otherwise, assess how the candidate evaluates alternatives, implements a reliable workflow, and adapts when models or requirements change.

From a real search

Do not turn adjacent experience into confirmed fit

During a Refery search, example profiles were shared for calibration with their fit rationale and interest status kept separate. Later candidate recommendations named both relevant work and questions the hiring team should still test.

For an AI engineer, use the same format to distinguish evidence of building an application from evidence of evaluating a model or operating the system. This is a lesson from the documented process, not a claim that every candidate was assessed through an identical technical exercise.

The useful lesson

Write down what the project demonstrates, what it does not establish, and the next evidence question.

Read the anonymized search story →