Edge ML Quantization Architect - Staff Engineer

Qualcomm; Santa Clara (CA); On-site; USD 161,000 - 241,000; Full time

Job summary

Qualcomm Technologies, Inc. is seeking a Staff Software Engineer to lead quantization and optimization of ML models for edge devices within the AI Hub ecosystem.

You will design and implement quantization pipelines (PTQ, QAT, mixed-precision) and develop tooling to deploy models on Snapdragon and other edge SoCs. You will collaborate across teams to integrate AIMET workflows with PyTorch, ONNX, and other frameworks, drive end-to-end quantization, and ensure accuracy, latency, and power targets

Qualifications

• BS in Computer Science, Engineering, Information Systems, or related field with 4+ years in HW/Software/Systems Eng.
• MS in CS, Engineering, Information Systems, or related field with 3+ years in HW/Software/Systems Eng.
• PhD in CS/Engineering/Information Systems or related field with 2+ years in HW/Software/Systems Eng.

Responsibilities

• Design, develop, and maintain quantization algorithms and compression pipelines within the AIMET framework (PTQ, QAT, mixed-precision, AdaScale).
• Implement weight-only/activation/KV-cache/sub-4-bit quantization for LLMs and AI models; quantify and mitigate accuracy degradation.
• Build tooling to analyze, profile, and debug quantization effects on model accuracy.

Skills

Python

PyTorch

ONNX

TensorFlow

Quantization

C++

Edge devices

Model optimization

Git

Education

Bachelor's degree

Master's degree

PhD

Tools

AIMET

GPTQ

AWQ

SmoothQuant

Job description

Qualcomm Technologies, Inc. is seeking a Staff Software Engineer to lead quantization and optimization of ML models for edge devices within the AI Hub ecosystem.

You will design and implement quantization pipelines (PTQ, QAT, mixed-precision) and develop tooling to deploy models on Snapdragon and other edge SoCs. You will collaborate across teams to integrate AIMET workflows with PyTorch, ONNX, and other frameworks, drive end-to-end quantization, and ensure accuracy, latency, and power targets