Machine Learning Engineer | Python | Pytorch | Distributed Training | Optimisation | GPU | Hybrid, San Jose, CA

• Productize and optimize models from Research into reliable, performant, and cost-efficient services with clear SLOs (latency, availability, cost).
• Scale training across nodes/GPUs (DDP/FSDP/ZeRO, pipeline/tensor parallelism) and own throughput/time-to-train using profiling and optimization.
• Implement model-efficiency techniques (quantization, distillation, pruning, KV-cache, Flash Attention) for training and inference without materially degrading quality.
• Build and maintain model-serving systems (vLLM/Triton/TGI/ONNX/TensorRT/AITemplate) with batching, streaming, caching, and memory management.
• Integrate with vector/feature stores and data pipelines (FAISS/Milvus/Pinecone/pgvector; Parquet/Delta) as needed for production.
• Define and track performance and cost KPIs; run continuous improvement loops and capacity planning.
• Partner with ML Ops on CI/CD, telemetry/observability, model registries; partner with Scientists on reproducible handoffs and evaluations.

• Bachelors in computer science, Electrical/Computer Engineering, or a related field required; Master's preferred (or equivalent industry experience).
• Strong systems/ML engineering with exposure to distributed training and inference optimization.

Industry Experience:

• 3-5 years in ML/AI engineering roles owning training and/or serving in production at scale.
• Demonstrated success delivering high-throughput, low-latency ML services with reliability and cost improvements.
• Experience collaborating across Research, Platform/Infra, Data, and Product functions.

Technical Skills:

• Familiarity with deep learning frameworks: PyTorch (primary), TensorFlow.
• Exposure to large model training techniques (DDP, FSDP, ZeRO, pipeline/tensor parallelism); distributed training experience a plus
• Optimization: experience profiling and optimizing code execution and model inference: (PTQ/QAT/AWQ/GPTQ), pruning, distillation, KV-cache optimization, Flash Attention
• Scalable serving: autoscaling, load balancing, streaming, batching, caching; collaboration with platform engineers.
• Data & storage: SQL/NoSQL, vector stores (FAISS/Milvus/Pinecone/pgvector), Parquet/Delta, object stores.
• Write performant, maintainable code
• Understanding of the full ML lifecycle: data collection, model training, deployment, inference, optimization, and evaluation.