Member of Technical Staff, AI-Driven Compilation

San Francisco Tensor Company•San Francisco, CA

1d•$275,000 - $315,000•Onsite

About The Position

At SF Tensor, we are building the future of high-performance compute by rethinking and rebuilding the AI stack from the hardware to the compiler to the cloud. Our goal is to eliminate the friction that slows down AI research and development, making compute faster, cheaper, and more accessible. We are developing a Kernel Optimizer that automatically finds the fastest possible form of code for any vendor and cluster topology, and a Model Foundry to manage runs, simplify research, and enable workload portability across clouds and chips. We are backed by prominent investors and are seeking individuals who believe that advancements in AI require corresponding leaps in compute capabilities.

Requirements

Strong background in reinforcement learning with hands-on experience training agents.
Experience building LLM agents, tool use and other agentic systems.
Familiarity in GPU programming concepts and a willingness to get down to the ISA.
Proficient in PyTorch or JAX.
Ability to design and run experiments that produce trustworthy results.

Nice To Haves

Experience in ML compiler stacks (XLA, TVM, Triton, MLIR) or LLVM backends.
Background in program synthesis, superoptimization, combinatorial search, formal methods, SMT solvers or verified compilation.
Familiarity with RLHF, reward modeling or preference learning.
Research contributions in RL or learned optimization.
Familiarity with GPU performance optimization, profiling and microbenchmarking.

Responsibilities

Design and implement RL systems that search over a massive program space: instruction selection, schedules, tile sizes, fusion strategies and phase ordering.
Build agentic compilation loops that use LLMs to reason about IR, propose transformations and learn from measured results.
Design the search and credit-assignment machinery around an exact reward built from measured latency on real hardware and a formal correctness proof.
Create representations and embeddings of compiler IR that hold up under learned optimization.
Build the training infrastructure for compiler optimization agents, including rollout throughput and distributed evaluation on real silicon.
Push transfer and cold-start performance so that the search works on unfamiliar targets and new ISAs from the first trial.
Close the loop with production workloads so that the compiler keeps improving from what customers actually run.
Work directly with our compiler and kernel engineers to land learned components in the shipping pipeline.
Run rigorous experiments, dig into the results, iterate on them and publish or open-source some of them.