Research Engineer - Inference

The role

This research engineering position sits within the research group and focuses on getting advanced models into live service. The person owns the pipeline from a trained checkpoint through to serving infrastructure, with an emphasis on speed, reliability, and cost for real-time streaming use.

What you would do

• Move newly trained models into production, owning the handoff from checkpoint to serving stack
• Improve inference speed, throughput, and cost using compression, distillation, cache tuning, batching, and custom kernels
• Build and tune low-latency serving systems for streaming, real-time traffic
• Develop internal tooling so researchers can release models quickly and with confidence in performance

What they asked for

• Hands-on experience running machine learning models in production, especially latency-sensitive services
• Solid GPU programming and inference optimization skills with tools like CUDA, Triton, TensorRT, or vLLM/SGLang
• Ability to independently find and remove bottlenecks across model, kernel, and orchestration layers
• Ability to build instrumentation for measuring performance
• Formal degrees or certifications are not required; portfolio work or open-source contributions are valued