This role is for one of our clients
Compensation: $90 - $120 per hour
Location: Remote
Job Overview
We are seeking an experienced ML Infrastructure and Performance Engineer to contribute to the development and advancement of next-generation Generative AI systems. This role is focused on building, optimizing, debugging, and evaluating the infrastructure that powers large language models and other frontier AI systems.
The ideal candidate will have hands-on experience in ML systems engineering, accelerator performance optimization, distributed workloads, or large-scale model serving. Expertise in areas such as custom accelerator kernels, performance profiling, trace analysis, distributed systems debugging, and high-throughput inference is particularly valuable. Experience with JAX, XLA, and TPU-based infrastructure will be a strong advantage.
In this role, you will apply your engineering expertise to create, evaluate, and improve technical tasks and solutions used to train and assess advanced AI models. Your work will help improve how frontier AI systems understand and solve complex ML infrastructure and performance engineering problems.
This is a fully remote opportunity open to candidates worldwide. The engagement will be structured in accordance with applicable local employment and compliance requirements. Candidates must be available for a full-time commitment of 40 hours per week, with no conflicting professional engagements.
Requirements
Key Responsibilities
- Design challenging technical tasks covering accelerator kernel development, performance profiling, distributed workload debugging, and high-throughput inference serving.
- Develop complex and adversarial engineering problems that test the capabilities of advanced AI models, including tasks involving accelerator programming and optimization.
- Create accurate, technically sound reference solutions that demonstrate best practices in ML systems engineering.
- Evaluate AI-generated solutions for correctness, efficiency, scalability, and practical applicability.
- Develop detailed technical guidelines and assessment rubrics for kernel optimization, profiler interpretation, distributed systems, and inference performance.
- Analyze performance bottlenecks across ML infrastructure, compilers, accelerators, and model-serving systems.
- Provide clear, structured technical feedback to help improve AI model capabilities in ML systems and infrastructure engineering.
- Collaborate with researchers and fellow subject matter experts to maintain the quality, consistency, and accuracy of AI training and evaluation data.
Required Qualifications
- At least 2 years of hands-on professional experience in ML systems, ML infrastructure, model serving, accelerator performance engineering, or a closely related field.
- Practical experience in at least one of the following areas:
- Accelerator Kernel Programming: Developing or optimizing custom kernels using Pallas, Triton, CUDA, CUTLASS, or similar technologies.
- Performance Profiling: Working with profiling and trace-analysis tools such as JAX/XLA Profiler, Kineto, torch.profiler, or NVIDIA Nsight.
- Distributed Workload Debugging: Identifying and resolving performance and execution issues in distributed or accelerator-intensive workloads.
- LLM Inference Serving: Experience with vLLM, SGLang, TensorRT-LLM, JetStream, KV cache optimization, paged attention, or continuous batching.
- Production-level experience with JAX and/or PyTorch.
- Understanding of modern accelerator architectures, including TPUs (v4–v6e), NVIDIA H100/B200, AWS Trainium, or AMD MI300.
- Ability to analyze and optimize throughput, latency, memory utilization, and computational efficiency.
- Demonstrable professional growth and career progression in engineering or infrastructure-focused roles.
- Availability to commit at least 40 hours per week on weekdays.
- Excellent written communication skills, with the ability to explain complex technical concepts and engineering decisions clearly.
Preferred Qualifications
- Experience with JAX, XLA, and TPU-based development environments.
- Knowledge of custom operators, sharding, and distributed training frameworks such as pjit, shard_map, FSDP, Megatron, or MaxText.
- Familiarity with compiler and graph-level technologies such as XLA, StableHLO, torch.compile, and MLIR.
- Experience developing performance benchmarks, technical evaluation frameworks, or engineering assessment rubrics.
- Exposure to AI model evaluation, synthetic training data generation, or working with frontier AI systems.
Engagement Details
- Work Arrangement: Fully Remote
- Geographic Eligibility: Worldwide
- Commitment: Full-Time, 40 hours per week
- Schedule: Weekdays
- Engagement Structure: Employment or engagement through an appropriate local partner, depending on the candidate's country of residence.
- Availability: Candidates must be able to commit exclusively to this engagement without conflicting professional commitments.