Spend panel hours on people worth the hour.
The people exist and are findable. Mira screens for fit and interest, then books first interviews with the evidence attached.
A Staff ML Engineer brief goes in. Metix AI runs the search. Booked interviews come back on a calendar. I weighted people who have written kernels and owned a training stack, not people who have only called a hosted API. CUDA was pass-fail. The technical interview stays with your engineers. 156 cleared the bar and were contacted. I ranked the 6 who came back so panel hours go to deciding, not to discovering a mismatch live.
Staff ML engineer. Booked interviews out.
I weighted people who have written kernels and owned a training stack, not people who have only called a hosted API. CUDA was pass-fail. The technical interview stays with your engineers.
Report figures are published under /reports/mapping/. Retrieval numbers are the Mira-Embeddings-V1 paper. Neither replaces your technical interview.
Teams already hiring through Metix AI: MiniMax, Dyna Robotics, R3D3 Ventures, Snaplii, SagaLab.
One anonymized account · two rounds
They came back and ran it again
Metix AI contacted 1,412 unique people for this account. The bars show how far those conversations went.
Read the full customer case →- People contacted
- 1,412
- Said they were interested
- 101
- Resumes received
- 50
Company records through 12 August 2026. Each bar is a share of the people contacted. Interview booking was not switched on for this account, so its record stops at resumes. This is one account, not a forecast for your role.
Supply is mappable. Panel hours are the constraint.
Every serious evaluation costs senior engineering time, so the funnel gets rationed to what a panel can afford.
Mira books the first interview, and you keep the bar.
- 01 You
Write the engineering bar
Stack, seniority, domain, and the work. Mira asks a follow-up only when the answer would change the search.
- 02 Mira, after you approve
Sourcing and outreach
900M+ profiles, not limited to one network. Nothing goes out under your name until you sign off.
- 03 Mira
Screening and booking
Mira screens replies for interest and fit before booking, so panel hours are not spent on misses.
- 04 You
Technical interview
Nothing we ship evaluates whether someone can build your system. That judgment stays with your engineers.
A recruiter who wants hands on every query runs Fast Search and Deep Search in the same plan. How Fast Search and Deep Search work.
The retrieval study: Mira-Embeddings-V1
The shortlist, the outreach, and the interview.
Hi Wei, Staff ML at a compute company. CUDA and PyTorch are must-haves. The panel still does the technical interview…
Needs your approvalProduct walkthrough for a sample role: Staff ML Engineer at Atlas Compute. Demo data, not a customer result.
See it run on your own role.
A 15-minute demo, no preparation needed. Or start the free 14-day trial with no card.
Start free. Plans from $99 a month.
No card required. 500 Credits cover 5 Deep Search lists, and running a Fast Search uses 0 Credits.
Start with the hardest role →Credits meter usage: 10 per email unlock, 20 per phone-number unlock, 100 per Deep Search list, and 3,000 per additional AI Agent above the plan allowance. Work inside an active Agent's workflow uses 0 Credits. Full plans and annual prices.
Where this fits, and where it does not.
A good fit
- Staff and senior IC roles, where a miss is expensive in senior hours.
- ML and research-adjacent roles. You still run the research interview.
- Infra, systems, and rare stacks. Sourcing is not limited to people active on one network.
Not built for
- Replacing your technical interview. Nothing we ship evaluates whether someone can build your system under your constraints.
- A candidate export. The published reports are talent-structure studies.
The longer version
Public reports on this site profiled 3,041 OpenAI engineers and 1,281 Anthropic engineers and researchers as full populations. The compute study found that about one in seven NVIDIA engineers in its U.S. sample (14.6%) have résumés spanning both silicon and ML systems, and that 44% have spent more than four years in their current role. Roles stay open because funnels get rationed to the width a panel can afford.
Before you start.
Can Metix AI evaluate specialized engineering skills?
Mira screens against the hiring bar you approve: required skills, seniority, domain background and the questions you would ask in a first call. The published Mira-Embeddings-V1 study covers how retrieval is tuned on recruitment data specifically. What Metix AI does not do is replace your technical interview; it ships the screening evidence with each booked candidate so your engineers spend their hours on people already worth the hour.
If entire engineering orgs can be mapped, why are roles still open?
The people exist. Public Metix AI reports profiled 3,041 OpenAI engineers and 1,281 Anthropic engineers and researchers as full populations, and 56,861 engineering and research professionals across 15 U.S. AI-compute and chip companies. In the AI-compute study's U.S. sample, about one in seven NVIDIA engineers (14.6%) are identifiable as dual-stack, with experience in both silicon and ML systems, and 44% have been in their current NVIDIA role for more than four years. The report reads that tenure, together with a roughly 10x stock run, as golden handcuffs. That reading is a proxy drawn from tenure and public stock moves; the report has no pay data. Roles stay open because every serious evaluation costs senior engineer hours, so funnels get rationed to the width a panel can afford.
Where do engineering candidates come from?
Sourcing runs across 900M+ profiles in more than 190 countries, per the published funding announcement. It is not limited to people active on one network, which matters for engineering searches where the strongest candidates are rarely applying anywhere.
Does Metix AI replace the technical interview?
No. Nothing we ship evaluates whether someone can build your system under your constraints. That judgment stays with your engineers. What arrives is a booked first interview with the screening evidence attached, so panel hours go to deciding rather than discovering a mismatch live.
Other desks and markets
Fix the constraint,
not the haystack.
The people are findable. Spend senior hours on people already worth the hour.