Staff Engineer — AI-Driven GPU Compiler Optimization

SF Tensor; San Francisco (CA); On-site; USD 275,000 - 315,000; Full time

Benefits offered by this job

Relocation assistance

Equity and benefits

Office in San Francisco

Job summary

SF Tensor in San Francisco is building the fastest GPU compiler and a suite of AI-driven tools. We’re hiring a Member of Technical Staff for AI-Driven Compilation to own the search, with a focus on reinforcement learning, LLM-enabled reasoning about IR, and scalable evaluation on real silicon.

You’ll work on agentic systems, integrate with compiler and kernel teams, and ship learned components into production. The role includes relocation assistance and competitive equity.

Qualifications

• Experience with reinforcement learning in practical systems.
• Experience building LLM-driven agents and tool use.
• Familiarity with GPU programming concepts and ISA-level work.
• Proficiency in PyTorch or JAX.
• Exposure to ML compiler stacks or LLVM backends is a plus.

Responsibilities

• Design and implement RL systems that explore a large program space (instruction selection, scheduling, tiling).
• Build agentic compilation loops using LLMs to reason about IR and learn from measurements.
• Define search and credit-assignment mechanisms based on latency and formal proofs of correctness.
• Create representations/embeddings of compiler IR suitable for learned optimization.
• Develop training infrastructure for compiler-optimization agents including rollout throughput and distributed evaluation.
• Ensure transfer and cold-start performance across unfamiliar targets and ISAs.

Skills

Reinforcement learning

LLM agents

GPU programming concepts

PyTorch

JAX

Tools

XLA

TVM

Triton

MLIR

LLVM backend

Job description

SF Tensor in San Francisco is building the fastest GPU compiler and a suite of AI-driven tools. We’re hiring a Member of Technical Staff for AI-Driven Compilation to own the search, with a focus on reinforcement learning, LLM-enabled reasoning about IR, and scalable evaluation on real silicon.

You’ll work on agentic systems, integrate with compiler and kernel teams, and ship learned components into production. The role includes relocation assistance and competitive equity.