AI Case Study 1

Compiler engineer for a global physical AI lab

Sector: Robotics foundation models

Stage: Seed (backed by Eclipse, Khosla Ventures, Bpifrance; angel investors include Eric Schmidt and Xavier Niel)

Location: SF / Paris / remote

Role: ML Compiler and Systems Engineer

The brief

The client was building an open-source robotics simulator and a universal robotics foundation model. They needed someone to own the compiler stack — JIT compilation, LLVM IR, GPU codegen — that made physics simulation and foundation model training fast enough for real iteration. The candidate pool was globally small: realistically, a few hundred engineers worldwide with the right mix of compiler construction, GPU performance, and ML framework experience.

The challenge

The client’s team already included the creator of the original PyTorch, HuggingFace infra contributors, and a Mistral multimodal lead. The bar was “can meaningfully improve the stack these people built.” Standard recruiting channels are useless at that level.

What we did

• Mapped the full target list: NVIDIA Warp and Triton teams, Meta PyTorch core and Inductor, Google XLA and JAX, Modular, OctoML, Groq, Cerebras, OpenAI Triton contributors, HuggingFace infra
• Sourced from OSS commit history on torch, Triton, MLIR, and the client’s own simulator
• Identified an MTS from a frontier lab already operating at the compiler + CUDA + Triton layer as the top pick within two weeks
• Cross-referenced four candidates already in our pipeline from adjacent searches (a Meta Sr. Staff graphics optimization engineer, a staff CUDA engineer at a surgical robotics company, an HPC-heavy PhD)

Outcome

The client moved into interviews with the shortlist ahead of their planned product release. The top candidate was already known to us from earlier searches at adjacent frontier AI companies.