Research Scientist / ML Engineer - Performance Engineering

About the role

The Biological Computing Co. is developing AI systems that combine generative models with biological computing and large-scale infrastructure. We need an optimization expert to refine our world-model and neural-optimizer projects by increasing the speed, throughput, and reliability of our production systems.

Key facts

What you'll do

• Refine inference performance for LLMs, diffusion models, video generation, and world-model systems.
• Manage serving efficiency using quantization, distillation, speculative decoding, KV caching, and memory optimization.
• Develop high-throughput inference pipelines for GPU clusters.
• Analyze model performance metrics including latency, memory usage, cost, and GPU utilization.
• Write custom kernels and low-level optimizations using Triton, CUDA, or PyTorch.
• Enhance distributed training and fine-tuning efficiency through parallelism, checkpointing, and data loading strategies.
• Collaborate with research teams to remove bottlenecks in model architecture and deployment workflows.
• Document performance gains as technical benchmarks for customer-facing applications.
• Balance the trade-offs between model quality, deployment costs, and inference speed.

Requirements

• Proven background in ML systems, high-performance infrastructure, or model optimization.
• Practical experience improving LLMs, diffusion models, or video generation systems.
• Proficiency in PyTorch and working at the intersection of model code and runtime environments.
• Ability to manage the balance between system performance and model output quality.
• Experience navigating ambiguous, early-stage project environments.

Nice to have

• PhD or MS in Computer Science, Machine Learning, Systems, Robotics, or a related field.
• History of optimizing large-scale generative models in research or production.
• Familiarity with stacks such as Triton, CUDA, vLLM, TensorRT, DeepSpeed, FSDP, or Ray.
• Experience specifically with world models.

Skills & tools

• PyTorch
• Triton
• CUDA
• Inference optimization (KV caching, quantization, pruning, distillation)
• Distributed training
• GPU profiling and debugging

Practical notes

• This role requires the ability to transition research prototypes into stable systems capable of supporting partner use cases and product demos.