SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Normal Computing is an applied AI company building foundational hardware and software for the semiconductor industry and critical AI infrastructure. The company operates across New York City, Silicon Valley (Mountain View), London, Copenhagen, and Seoul.
As a Compiler Frontend Engineer, you will design and extend the compiler infrastructure that enables new large language model (LLM) architectures to run on Normal's specialized hardware. Normal's hardware executes memory-bound inference operations directly in memory rather than shuttling weights to a processor, and the compiler is the critical bridge between models and this novel architecture.
Your primary responsibilities include:
- Model Bring-up: Support new LLM architectures in the compiler frontend, from attention variants to mixture-of-experts (MoE) routing schemes, ensuring they run correctly on the stack soon after release.
- IR and Lowering: Extend the intermediate representation (IR), operations, and lowering passes to faithfully represent new architectural primitives with comprehensive test coverage.
- Quantization and Numerics: Implement support for quantization schemes and numeric formats used by new models, ensuring lowering preserves model accuracy.
- Architecture Tracking: Monitor the LLM architecture landscape to recommend which architectures the team should prioritize.
- Cross-Team Collaboration: Hand off lowered programs to backend compiler engineers and work with the LLM serving engineer to move models from compiler support to production serving.
This is a research-adjacent role where you will follow compiler and ML literature closely, potentially publishing on behalf of Normal. You will validate bring-up against simulators and FPGA prototypes before silicon arrives, ensuring models are production-ready.
REQUIREMENTS:
- Hands-on ML compiler experience with shipped compiler passes
- Deep familiarity with PyTorch internals, including compilation stack (TorchDynamo, FX, torch.export)
- Working knowledge of modern LLM architectures: attention mechanisms, MoE routing, quantization schemes
- Patience for detailed work: progressing architectures from "runs" to "runs correctly" to "runs correctly and fast"
- Comfort designing compiler infrastructure in young codebases and extending established ones
- Comfort building software for evolving hardware, from simulation through bring-up
- Fluency in Python; familiarity with C++ or Rust
BONUS QUALIFICATIONS:
- Experience with ML compilers (IREE, XLA, TVM)
- Familiarity with JAX, ONNX, or other ML frameworks
- Model partitioning experience (tensor, pipeline, or expert parallelism)
- Hardware-software co-design experience
- Non-standard or in-memory hardware target experience
- Open-source compiler or ML infrastructure contributions
