Job description

We are looking for a Senior Inference Engineer to own inference for real-time multimodal conversational AI. This is a full-stack inference role: you will work across the entire path a model takes from research to production — shaping model architecture so it is servable, building the real-time runtime that serves it within hard latency budgets, and building the offline systems that train and reinforce it. You will operate at the boundary of Science and Inference, taking frontier-scale speech and audio models and making them run within real-time latency budgets on production hardware. You will co-design architectures with scientists to make them inference-friendly from inception, own the low-latency streaming serving path, and build the training and reinforcement-learning infrastructure that closes the loop. You will have the compute, data, and runway to solve problems that few teams in the world are positioned to tackle. As a Senior Engineer, you will own a significant area of the inference stack end to end, drive its technical execution, contribute to the team's roadmap, and work closely with scientists and hardware partners to ensure our models run fast enough to feel human in real time — and at a cost that makes them viable at scale. You may go deep in one of the areas below while contributing across the others. Key job responsibilities Model Architecture & Inference Co-Design • Partner with research scientists to make model architectures servable from inception — surfacing the latency, memory, and cost implications of architecture choices before they are locked in • Implement and optimize the inference path for large-scale multimodal models — attention and KV-cache mechanisms, multimodal/autoregressive decoding, and the compute primitives on the critical path Apply efficiency techniques across the stack — quantization (per-tensor/per-channel/per- group, INT8/FP8/BF16), speculative decoding, operator fusion, and paged KV-cache — and quantify their quality/latency trade-offs • Develop and tune high-performance kernels for critical operations where off-the-shelf implementations leave performance on the table, integrating them into production serving with minimal overhead • Profile end-to-end performance with tools such as Nsight Compute/Systems and roofline analysis to identify and eliminate bottlenecks in large-scale inference workloads Real-Time & Interactive Runtime • Own the real-time serving path for streaming multimodal conversational AI, meeting sub- second, streaming latency budgets under concurrent session load • Build and tune continuous batching, scheduling, and preemption to balance throughput against per-request latency SLAs for interactive workloads • Customize production serving frameworks (e.g., vLLM, PyTorch) for real-time streaming generative models that fall outside standard LLM serving patterns — sustained low-latency output under concurrent session load • Implement multi-GPU inference (tensor parallelism, collective communication) for latency- critical paths, and drive cost toward parity with existing production baselines • Establish latency, throughput, and cost benchmarking, and publish the operational metrics that gate deployment Offline Systems: Training, RL & Evaluation Infrastructure • Build and scale the offline inference systems behind post-training — high-throughput rollout generation and reward-model serving for reinforcement learning (RL/RLHF/RLAIF) • Ensure train/serve consistency — that the inference path used in RL and evaluation faithfully matches production online behavior (e.g., parity across sampling and logit processing) • Work with the evaluation team to enable offline inference that captures the quality dimensions unique to real-time conversation — latency sensitivity, audio quality, and interaction naturalness - 5+ years of non-internship professional software development experience - 5+ years of programming with at least one software programming language experience - 4+ years of leading design or architecture (design patterns, reliability and scaling) of new and existing systems experience - Bachelor's degree in computer science or equivalent - Experience as a mentor, tech lead or leading an engineering team - 2+ years of hands-on experience optimizing inference for neural models — not just using inference frameworks, but profiling and improving them - Strong understanding of deep learning architectures (transformers, attention mechanisms, autoregressive decoding) and their application to speech/audio or other multimodal domains - Production track record delivering latency-constrained, real-time inference systems under concurrent load - Experience with GPU performance optimization — memory hierarchy, occupancy, KV-cache management, and the accelerator programming model - Demonstrated ownership of a technical area — driving execution for a workstream and collaborating effectively across scientists and engineers - Experience with production LLM/multimodal serving internals (e.g., vLLM, TensorRT-LLM): scheduler, batching, block manager, sampler customization - Hands-on experience building real-time or streaming AI systems — speech, audio, or video — with hard latency budgets - Experience authoring custom GPU kernels (CUTLASS, Triton, raw CUDA/PTX), fused attention (FlashAttention-style), or quantized GEMM - Familiarity with model-compression and efficiency techniques — quantization, pruning, distillation, speculative decoding, long-context optimization - Experience building offline inference or rollout/reward-serving infrastructure for reinforcement learning or large-scale evaluation - Experience with distributed training and post-training pipelines (SFT through RL) — parallelism strategies, training stability, and multi-accelerator communication (NCCL, NVLink) - Familiarity with multiple hardware backends (NVIDIA GPU, AWS Neuron/Trainium, edge accelerators) and how architecture choices affect inference latency, memory, and cost - Background in speech-to-speech or audio generative models (codec models, autoregressive audio generation), speech recognition, or speech synthesis - Experience shipping research to production at scale — models serving real users, not just benchmark results - Contributions to open-source inference/kernel projects (vLLM, CUTLASS, FlashAttention, TensorRT-LLM, Triton, or similar) Amazon is an equal opportunity employer and does not discriminate on the basis of protected veteran status, disability, or other legally protected status. Los Angeles County applicants: Job duties for this position include: work safely and cooperatively with other employees, supervisors, and staff; adhere to standards of excellence despite stressful conditions; communicate effectively and respectfully with employees, supervisors, and staff to ensure exceptional customer service; and follow all federal, state, and local laws and Company policies. Criminal history may have a direct, adverse, and negative relationship with some of the material job duties of this position. These include the duties and responsibilities listed above, as well as the abilities to adhere to company policies, exercise sound judgment, effectively manage stress and work safely and respectfully with others, exhibit trustworthiness and professionalism, and safeguard business operations and the Company’s reputation. Pursuant to the Los Angeles County Fair Chance Ordinance, we will consider for employment qualified applicants with arrest and conviction records. Our inclusive culture empowers Amazonians to deliver the best results for our customers. If you have a disability and need a workplace accommodation or adjustment during the application and hiring process, including support for the interview or onboarding process, please visit amazon.jobs for more information. If the country/region you’re applying in isn’t listed, please contact your Recruiting Partner. The base salary range for this position is listed below. Your Amazon package will include sign-on payments and restricted stock units (RSUs). Final compensation will be determined based on factors including experience, qualifications, and location. Amazon also offers comprehensive benefits including health insurance (medical, dental, vision, prescription, Basic Life & AD&D insurance and option for Supplemental life plans, EAP, Mental Health Support, Medical Advice Line, Flexible Spending Accounts, Adoption and Surrogacy Reimbursement coverage), 401(k) matching, paid time off, and parental leave. Learn more about our benefits at amazon.jobs . USA, CA, Sunnyvale - 193,300.00 - 261,500.00 USD annually USA, MA, Boston - 168,100.00 - 227,400.00 USD annually USA, WA, Seattle - 168,100.00 - 227,400.00 USD annually

Requirements

Must-haves

We are looking for a Senior Inference Engineer to own inference for real-time multimodal
conversational AI. This is a full-stack inference role: you will work across the entire path a model
takes from research to production — shaping model architecture so it is servable, building the
real-time runtime that serves it within hard latency budgets, and building the offline systems
that train and reinforce it.
You will operate at the boundary of Science and Inference, taking frontier-scale speech and
audio models and making them run within real-time latency budgets on production hardware.
You will co-design architectures with scientists to make them inference-friendly from inception,
own the low-latency streaming serving path, and build the training and reinforcement-learning
infrastructure that closes the loop. You will have the compute, data, and runway to solve
problems that few teams in the world are positioned to tackle.
As a Senior Engineer, you will own a significant area of the inference stack end to end, drive its
technical execution, contribute to the team's roadmap, and work closely with scientists and
hardware partners to ensure our models run fast enough to feel human in real time — and at a
cost that makes them viable at scale. You may go deep in one of the areas below while
contributing across the others.
Key job responsibilities
Model Architecture & Inference Co-Design
Partner with research scientists to make model architectures servable from inception —
surfacing the latency, memory, and cost implications of architecture choices before they are
Implement and optimize the inference path for large-scale multimodal models — attention
and KV-cache mechanisms, multimodal/autoregressive decoding, and the compute
primitives on the critical path
Apply efficiency techniques across the stack — quantization (per-tensor/per-channel/per-

Nice-to-haves

None flagged.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aigollmaws

Platform: amazon
Effort: medium

Benefits

Benefits detected in the listing.

Health Insurance

Dental

Vision

401(K)

Paid Time Off

Parental Leave