In-Memory AI Inference Algorithms Engineer

Aimlroles; Palo Alto (CA); On-site; USD 180,000 - 260,000; Full time

Job summary

Normal Computing is an applied AI company building the hardware and software foundations for in-memory AI inference. You will create algorithms that run on our memory-centric hardware, redefining how attention and long-context decoding map to this substrate.

This co-design role blends architecture decisions with algorithm development, testing in simulators and on silicon as hardware evolves. The ideal candidate combines deep large-model inference knowledge with mathematical insight and hands-on

Qualifications

• Strong understanding of large model inference and attention mechanisms.
• Experience optimizing inference: quantization, sparsity, kernel fusion, or memory-efficient attention.
• Track record of implementing algorithms close to hardware from theory to working hardware.

Responsibilities

• Develop algorithms for LLM inference on in-memory hardware.
• Collaborate with hardware/architecture teams to shape native computations.
• Design numerical methods that exploit hardware dynamics.
• Build evaluation frameworks and benchmarks for hardware or simulation.
• Translate model workloads into hardware design constraints and opportunities.
• Prototype and iterate rapidly as hardware evolves toward silicon.
• Optimize performance at both gate and algorithmic levels.

Skills

Large model inference

Attention mechanisms

Memory bandwidth

Inference optimization

Python programming

Systems programming

Cross-team collaboration

First principles reasoning

Education

PhD preferred in ML/Math/Physics/EE

Tools

FPGA

Job description

Normal Computing is an applied AI company building the hardware and software foundations for in-memory AI inference. You will create algorithms that run on our memory-centric hardware, redefining how attention and long-context decoding map to this substrate.

This co-design role blends architecture decisions with algorithm development, testing in simulators and on silicon as hardware evolves. The ideal candidate combines deep large-model inference knowledge with mathematical insight and hands-on