SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
AssemblyAI is seeking a Senior Research Engineer to join the Research team and improve the systems powering large-scale distributed training, data processing, and inference for Voice AI models. The company processes 1M+ hours of audio daily and serves 600M+ monthly inference calls, with thousands of customers including Granola, Fireflies, Figure AI, and CallRail.
In this role, you will raise experimental velocity by making it faster to launch experiments, get trustworthy results, and determine next steps. Key responsibilities include maintaining and evolving the JAX training framework for large-scale TPU distributed training, improving model training data by investigating quality issues and building tooling to surface them, analyzing production model accuracy and building evaluation harnesses, translating research prototypes into production-ready systems, optimizing production inference for speech language models using techniques like quantization and speculative decoding, and investigating performance bottlenecks across the stack from low-level kernels to high-level system design.
You will work cross-functionally with researchers, infrastructure teams, and production engineering—not as a handoff point, but as someone who learns enough of each domain to follow problems through to resolution. You'll train models, run evaluations, and analyze data yourself to deliver direct impact and identify what's worth building to multiply team output.
Required qualifications include expert-level proficiency with JAX and TPUs (including Flax, Optax, and the XLA compilation pipeline), strong measurement discipline with skepticism toward results until validated, deep understanding of modern deep learning systems, expertise in layer-level optimization and large-scale distributed training, knowledge of streaming and low-latency asynchronous inference, familiarity with inference compilers and advanced parallelization techniques, and the ability to understand end-to-end impact and prefer rapid iteration over extended planning cycles.
AssemblyAI operates as a lean, capital-efficient organization with under 100 people generating roughly $500K ARR per employee. The company emphasizes meritocracy, real ownership, and fast movement without heavy bureaucracy or approval processes.
