■ A note from the managing partner
Ines Varga — Managing Partner, London
The best engineers and researchers don’t apply for jobs. They get one real chance every few years to make a move, and they want that conversation to be with someone who understands the work: the systems, the strategies, the trade-offs.
That is the bar we hold ourselves to. We spent the last fifteen years inside the hiring rooms of AI labs, trading firms and the teams competing with both. We don’t pitch roles. We have honest conversations about where someone should spend the next five years, and we are often the reason it’s the right call.
— Ines Varga, for the partners
Senior placements since 2011
Across AI labs, systematic funds and deep-tech teams in nine countries.
Median days, brief to signed offer
Down from 84 in 2022, mostly because we cut the number of interview rounds we allow.
Still in role after two years
We re-run the search free if a hire leaves in the first year. We have done it four times.
First calls on a typical search
Every one made by a partner. No sourcers, no automated sequences, no spray-and-pray.
Practices
■ What we search for
Four practices, each run by a partner who has done the job or hired for it for a decade. We turn down about a third of the briefs we are offered.
01
AI & ML research
Research scientists and engineers who own a problem from paper to production. We know which labs are hiring, what the compute budget really is, and who has actually shipped.
02
Inference & infrastructure
The engineers who keep models and markets running: serving, GPU efficiency, data platforms and the on-call rota nobody wants to own.
03
Quant & trading
Researchers who own a signal end to end, and the engineers who make it tradable. Most of these searches start with a pod lead, not HR.
04
Leadership & team builds
Whole functions, not single seats: heads of engineering and research, first ten hires, and build-outs that run for years.
Searches
■ On the desk now
Every search we run is published here, open or closed, with the funnel as it stands. Drag to browse, or open one to read the brief.
#0412
Live
Head of Inference
Series C AI lab, 120 people
4,812
Mapped
212
Called
38
Met
6
Short
London
Week 5 of 9
#0409
Live
Quant Researcher, Futures
Systematic fund, $9bn AUM
1,904
Mapped
96
Called
21
Met
0
Short
Chicago
Week 3 of 8
#0407
Live
Founding ML Engineer
Seed-stage robotics company
820
Mapped
64
Called
19
Met
4
Short
Munich
Week 6 of 8
#0401
Closed
Head of Low-Latency Platform
Electronic market maker
2,210
Mapped
140
Called
31
Met
5
Short
Amsterdam
Placed in 54 days
#0396
Closed
VP Engineering
Series B developer-tools company
3,160
Mapped
180
Called
27
Met
6
Short
New York
Placed in 71 days
#0390
Closed
Research Scientist, Post-training
Frontier AI lab
1,280
Mapped
88
Called
24
Met
6
Short
Paris
Placed in 48 days
#0386
Closed
Head of Data Platform
Regulated digital-asset exchange
1,740
Mapped
110
Called
22
Met
5
Short
Singapore
Placed in 66 days
#0381
Closed
Chief Scientist
Climate-modelling startup
640
Mapped
52
Called
14
Met
4
Short
Zürich
Placed in 93 days
All searches, open and closed
→
Placements
■ Recent hires
Who we placed, where, and how long it took. Hover a row to see the team they joined.
2026
Inference & infrastructure
Head of Inference
Parallax Compute
Wasn’t looking. Moved for the serving roadmap, not the number.
61 days
2026
Quant & trading
Staff Quant Researcher
Stillwater Systematic
Second conversation was with the pod lead, not HR. That did it.
49 days
2025
AI & ML research
Founding Engineer
Halyard Robotics
Took a lower base for 1.6% and a lab bench of their own.
38 days
2025
Quant & trading
Head of Research Platform
Vellum Quant
Built the platform the researchers had been asking for since 2021.
57 days
2025
Leadership & team builds
VP Engineering
Orrery Labs
Doubled the team in eleven months; zero regretted attrition.
71 days
2025
AI & ML research
Principal ML Engineer
Northbeam AI
Left a frontier lab for a smaller one with a clearer shot at shipping.
44 days
Every placement since 2011
→
Method
■ Nine weeks, brief to signed offer
Every search runs the same way, and you can see every step of it. The Friday note shows the whole funnel: who we called, who said no and why.
How a search runs
→
Week 0
01 / 05
Brief
A two-hour session with the hiring manager and whoever the hire will work with most. We write the brief together and agree what would make you hire nobody.
Weeks 1–2
02 / 05
Map
We chart every team doing the work, name by name. For a typical search that is 3,000 to 5,000 people. You see the map before we call anyone.
Weeks 2–4
03 / 05
Approach
A partner makes every first call. We lead with the work, the team and the number, and we log every no and the reason for it.
Weeks 4–7
04 / 05
Meet & shortlist
Two-hour conversations with the strongest thirty to forty, then a shortlist of six with our notes, references we took ourselves and the offer that would move each one.
Weeks 7–9
05 / 05
Offer & close
We run the offer, the counter-offer and the notice period, then stay close for the first year. If the hire leaves in year one, we run the search again at no charge.
Voices
■ Clients and candidates
Half of our work comes from people we placed who later became the ones hiring. These are their words, lightly edited, names withheld.
“
Candidate
They talked me out of two offers before they found the right one. I didn’t know recruiters did that.
Staff ML engineer
Placed at an AI lab, 2025
“
Client
Six people on the shortlist and we wanted to hire four of them. That has never happened to us before.
CTO
Series C AI lab, London
“
Client
Keldwell knew our research stack better than half the people interviewing for it.
Head of Research
Systematic fund, Chicago
“
Candidate
They were honest that the equity was worth less than the startup said. I took the job anyway, with a better number.
Founding engineer
Robotics startup, Munich
“
Client
Nine weeks from brief to signed offer, for a role we had failed to fill for a year.
VP People
Developer-tools company, New York
“
Candidate
One call a quarter for three years, never a pitch. When I was ready, it was the first number I rang.
Research director
Placed in 2024
Partners
■ Who you’ll actually talk to
Small team, high signal. Every search is run from brief to offer by one of these five. No junior researchers cold-messaging on our behalf.
■ For candidates
Not looking? Good.
Most of the people we place weren’t looking when we first called. Tell us what would make a move worth it, and we’ll only call when something is. One conversation a quarter, never a pitch.
Talk to us in confidence
→
Confidential profile
Role I’d move for
Head of Inference
Number that would move me
£300k + meaningful equity
Only call me about
Series B+ labs · London or remote
Timing
Not before Q2 — bonus lands in March
Send my CV without asking
We never send a CV without your written permission. Ever.