Scaling large language model training throughput by forty percent at NVIDIA defines my professional focus on building resilient machine learning infrastructure. My recent work involved optimizing the distributed training stack for trillion parameter models, ensuring high availability across vast compute clusters. Together AI stands out as a leader in the decentralized cloud revolution, and my deep expertise in orchestrating large scale training pipelines aligns with the demands of building out your distributed inference and training platforms. This role provides a unique opportunity to apply my background in core infrastructure to your open Senior ML Infrastructure Engineer position.
During my tenure at Meta, I implemented PyTorch FSDP to reduce memory overhead by thirty percent for dense embedding layers during training runs. At Databricks, I configured complex Kubernetes clusters on AWS EKS using Terraform, enabling seamless scaling for data scientists deploying distributed jobs. While at Hugging Face, I optimized Tensor Parallelism workflows through custom Megatron-LM integration, which cut latency for real time inference tasks by twenty five percent. These experiences collectively refined my ability to manage the intricate lifecycle of machine learning infrastructure, focusing specifically on performance bottlenecks and cluster stability under high load.
Together AI represents the most ambitious approach to democratizing access to large models through your focus on low latency inference and optimized training backends. My specialized knowledge in distributed compute orchestration and hardware abstraction allows me to contribute immediately to your efforts in streamlining multi-node training clusters. I value your commitment to open research and the technical rigor required to sustain high throughput in competitive production environments. Combining my tactical experience with infrastructure automation and architectural strategy, I can help accelerate your roadmap for expanding the reach and capability of your core machine learning compute primitives.
My background uniquely prepares me to address the specific challenges involved in scaling infrastructure at Together AI. I possess the hands on technical depth to diagnose performance regressions in large clusters while designing robust automated workflows that empower engineering teams. I look forward to discussing how my experience with GPU resource scheduling and deep learning framework optimizations can translate into direct gains for your platform. Please let me know when you are available for a conversation regarding your current technical priorities and how my skills align with your upcoming infrastructure milestones.
Best regards,
Sofia Martinez
