AI Model Optimization Architect

About Qualcomm Cloud AI Team

Qualcomm leverages its expertise in compute, connectivity, and AI acceleration to drive innovations in Cloud AI. The Cloud AI team builds both hardware and software platforms to enable efficient inference of large-scale foundation models.

Role Overview

We are looking for a Staff Engineer specializing as an AI Model Optimization Architect who will lead comprehensive model transformation and optimization for large language models (LLMs), vision-language models (VLMs), diffusion models, and multimodal models running on Qualcomm’s inference accelerators. Collaboration with compiler, performance, and accuracy teams is crucial to adapting models for efficient accelerator execution balancing throughput, latency, memory use, and output quality. The role covers initial enablement through to production deployment, emphasizing scalable optimization for future hardware architectures.

Key Duties

• Develop and implement model optimization strategies to transform PyTorch-based models for efficient execution on Qualcomm accelerators.
• Manage graph capture and deployment workflows using PyTorch, ONNX, and torch.compile, including model rewriting and graph-level transformations.
• Create and optimize fusion kernels via domain specific languages like Triton, improving fused operations and rewriting performance-critical algorithms.
• Collaborate closely with compiler, performance optimization, and accuracy teams to co-design lowering procedures, kernel fusion techniques, memory layout decisions, and runtime system integration.
• Analyze and enhance inference performance for LLMs, VLMs, and diffusion models targeting throughput and latency across various batch sizes, sequence lengths, and serving modalities.
• Lead transformer-specific performance enhancements including management of KV cache, decoding procedures, and optimization for extended context lengths.
• Implement and optimize continuous batching mechanisms, understanding their effects on memory usage, scheduling dynamics, and tail latency.
• Design and scale distributed inference strategies such as sharding and parallel processing across multicore and multi-device platforms.
• Develop scalable, reusable methods and tooling to apply model optimizations effectively across new Qualcomm hardware architectures.
• Investigate and resolve advanced performance or stability issues ensuring production quality solutions.

Required Qualifications and Skills

• Expert proficiency with PyTorch and strong Python engineering focused on inference optimization.
• Hands-on experience using torch.compile, TorchDynamo, or comparable graph capture and compilation tools.
• In-depth knowledge of transformer architectures, including attention mechanisms, mixtures of experts (MoEs), and related performance trade-offs.
• Practical understanding of KV cache behaviors, serving time optimizations, and related performance vs. memory trade-offs.
• Solid background in computer architecture, machine learning accelerators, and distributed systems design.
• Proven experience leading cross-team technical projects and influencing architectural decisions.
• Advanced degree (MS or equivalent) in Computer Science, Machine Learning, Computer Engineering, Electrical Engineering, or a closely related technical discipline.

Preferred Qualifications

• Experience in building fusion kernels using DSLs like Triton, including working with ML compiler teams.
• Familiarity with LLM serving infrastructure and continuous batching frameworks.
• Background in numerical methods, and expertise in balancing performance and accuracy or building evaluation frameworks.

Location Advantages

Cork, as Ireland's second largest economic hub, hosts over 170 multinational companies with a workforce of nearly 39,000. The region enjoys cultural diversity and a high quality of life. Cork Airport connects to nearly 50 international destinations, including transatlantic routes.

Equal Employment Opportunity

Qualcomm is an inclusive employer committed to offering equal opportunity regardless of race, color, religion, gender identity, sexual orientation, national origin, disability, veteran status, or any protected category.

Employee Benefits

• Competitive salary, equity stock options, and performance-based bonuses
• Generous maternity and paternity leave policies
• Employee stock purchase plans
• Retirement pension matching programs
• Educational assistance for professional growth
• Support for relocation and immigration where applicable
• Comprehensive life, medical, income, and travel insurance coverage
• Subsidized memberships promoting physical and mental wellness
• Bicycle purchase assistance programs
• Access to employee-run clubs including running, football, chess, badminton, among others

Minimum Qualifications

• Bachelor’s degree in Engineering, Information Systems, Computer Science, or related field with at least 6 years of software engineering or equivalent experience.
• OR Master’s degree with minimum 5 years of relevant software engineering experience.
• OR PhD with minimum 4 years of relevant software engineering practice.
• Minimum 3 years of professional programming experience with languages like C, C++, Java, or Python.
• Equivalent experience will be evaluated on demonstrated abilities fulfilling the role’s responsibilities.

Additional Information

Qualified individuals with disabilities may request accommodations during the hiring process. Qualcomm is dedicated to accessibility and will provide necessary support.

Employees are expected to comply with company policies including information security and confidentiality obligations.

Recruitment agencies and external contractors are requested not to submit unsolicited applications or resumes.