Baseten builds AI infrastructure that powers production inference for some of the fastest-growing companies in the world. When they needed to scale their Forward Deployed Engineering team, they needed engineers with the depth to optimize GPU-level inference — and the communication skills to own enterprise customer relationships.
Backed by IVP · BOND · Greylock · Conviction · Altimeter · Battery Ventures · NVIDIA · 01 Advisors · BoxGroup · Blackbird Ventures
The Mandate
Baseten's FDE team is a full engineering function — one of four core engineering teams at the company. FDEs contribute to inference optimization, latency reduction, and production infrastructure. They just happen to work directly with customers while doing it.
FDEs work across LLMs, inference systems, benchmarking, diffusion models, and GPU optimization — touching the full surface area of what customers need. The role evolves: engineers self-select into specializations (performance, integrations, model optimization) over time.
“The product should eventually close every gap FDE is filling today. But customer need in AI is accelerating so fast that product can barely keep up. The need for FDE's to drive AI outcomes is growing, not shrinking.”
The Challenge
Any one of these alone would compress a funnel. All three together is why over 6,000 candidates produced eight hires.
The moment a top engineer hears the words customer-facing, they disengage. The pitch had to reframe the role entirely — not as customer support, but as operating at the tip of the spear between product capability and what the best AI companies in the world are trying to build. Most never got that far.
Engineers who passed filter one still needed something uncommon: the ability to take a production infrastructure problem they'd never seen before and explain it clearly to an enterprise customer in real time. Not polished. Not scripted. Actually thinking out loud at expert level.
Baseten's technical interviews used real production problems — KV cache design, inference optimization, throughput constraints. The bar wasn't algorithms; it was structured thinking and production-grade code discipline. Even strong engineers with the right background often didn't make it through.
The Approach
Before scaling volume, we ran a hypothesis-driven calibration across four candidate profiles. Each had a logical case. Only one held up against Baseten's technical bar.
Archetype 1
Hypothesis
Natural fit — already comfortable with customers, familiar with technical sales and deployment.
Reality
Strong communicators. Failed the technical screen at a high rate. The work they'd done was demos and integrations, not production systems.
Archetype 2
Hypothesis
Domain-aligned — work with models, understand inference, familiar with the AI stack.
Reality
Many ML engineers build prototype-level code in notebooks rather than production systems. They lacked the computer science fundamentals the technical screen was testing for.
Archetype 3
Hypothesis
Deep technical knowledge — especially those coming from applied research roles at top labs.
Reality
Research skills don't map directly to production code discipline. Passed conceptually; struggled to write structured, tested code under pressure.
Archetype 4
Hypothesis
The deepest engineering fundamentals. Just needs the right framing to want a customer-facing role.
Reality
Highest technical pass rate. The engineers who talked about why they were solving a problem — not just how — were the ones who converted. The best candidate we saw had zero AI keywords on her resume.
Execution
Once the winning profile was identified, reaching enough of them required retooling how we operate.
We ran five distinct sourcing lanes simultaneously — not sequentially. High-performance inference engineers, ML engineers at scale-ups, elite SWEs from top AI ecosystems, customer-facing engineers with production code depth, and former founders. Each lane had its own targeting logic, keyword strategy, and quality signals.
Baseten was growing fast enough that its customer list — which is our exclusion list — was expanding in near real-time. Companies that were valid targets one week became off-limits the next. We had to continuously update our sourcing infrastructure to account for it.
Job titles and keywords were unreliable. The differentiator was how engineers talked about their work — whether they described problems in terms of outcomes and constraints, or just tasks completed. M Search's screening layer was built to catch that distinction before the client's time was spent.
The Numbers
A 0.13% conversion rate - not a failure of process, but a reflection of exactly how rare this profile is.
Timeline
Sep 5
Search Kickoff
Sep 12
Market Mapping Delivered
Oct 6
First Shortlist Delivered
Nov 13
Final Client Interviews
Dec 9
Offer Extended
Outcomes
8
FDE hires
Engineers placed into the Forward Deployed team from elite AI infrastructure and software engineering backgrounds.
89%
offer acceptance rate
Candidates who reached the offer stage were well-qualified and well-prepared. The screening process reduced late-stage drop-off.
13 wks
time to first hire
From kickoff to signed offer — including full market mapping, calibration, multi-stage screening, and client interviews.
Engagement ongoing
Following the FDE engagement, M Search expanded the partnership to include an AI Support Engineering search — a deployed SRE function supporting Baseten's most strategic enterprise accounts. Additional hires are in progress.
From the Team
“We've worked with M Search on multiple searches, and what's consistent every time is how precisely they calibrate to our bar. They don't just understand the technical requirements — they understand the commercial instincts and domain context that determine whether someone will actually succeed in the role. Finding people who can hold their own across all three dimensions is genuinely rare, and M Search has demonstrated repeatedly that they know exactly how to find them. The rigor of their process and the quality of the slate they deliver is exceptional.”
“M Search came to us highly recommended, and working with them made it immediately clear why. Graham and his team operated as a true extension of our internal talent function — closely aligned with our hiring managers, deeply embedded in the requirements, and genuinely invested in getting it right. What set them apart wasn't just the quality of the candidates they delivered, but the depth of their understanding of what we actually needed: the technical bar, the communication style, the cultural fit. They didn't run a search process; they built a picture of the ideal person and then went and found them. I'd work with M Search again without hesitation.”
Work With M Search
M Search is a retained executive search firm focused on GTM and sales leadership for PE-backed and VC-backed B2B software companies. The Baseten engagement came from the same place all our work does: a mandate that required genuine judgment about rare talent.