AI Acceleration Algorithm Engineer

About the role We are looking for an AI Acceleration Algorithm Engineer to research, design, and productionize algorithms that improve end-to-end AI compute, retrieval, storage, and inference efficiency. The role emphasizes independent algorithmic innovation and measurable system-level benefits across accuracy, throughput, latency, memory capacity, and cost. What you will do - Design inference optimization algorithms for dynamic batching, request scheduling, parallelism, distributed sharding, load balancing, and elastic resource allocation. - Research model-efficiency techniques including post-training quantization, quantization-aware training, low-precision inference, structured sparsity, pruning, and distillation. - Develop high-performance retrieval, vector-search, data-placement, storage-caching, and I/O optimization algorithms for AI workloads. - Build cost models, search strategies, and adaptive optimization algorithms that balance accuracy, throughput, latency, memory capacity, and operating cost. - Productionize algorithms in training or inference systems and evaluate them using reproducible baselines, ablations, and end-to-end metrics. What we are looking for - Master’s degree or Ph.D. in Computer Science, AI, High-Performance Computing, Operations Research, or a related field. - Representative work in one primary area and meaningful experience in at least one adjacent area: retrieval, AI storage optimization, inference scheduling and acceleration, quantization, sparsification, or AI compute-platform optimization. - Evidence of an independent algorithmic contribution, rather than framework integration or parameter tuning alone. - Proficiency in C++/Python, Linux, CUDA or equivalent accelerator development, and mainstream frameworks such as PyTorch, TensorRT-LLM, vLLM, SGLang, or Megatron-LM. - Quantified results using metrics such as accuracy, throughput, TTFT, TPOT, latency, memory footprint, scalability, or cost. Preferred qualifications - Production deployment at significant model, traffic, data, or cluster scale. - Strong publications, patents, open-source contributions, or benchmark results in a relevant area.