Performance Engineer - Inference
ZenteiQ.aiBengaluru, IndiaOn-Site
Needs a free Superjobs account. You'll finish on ZenteiQ.ai's site.
Skill score
Summary
This role investigates and improves end-to-end performance of the inference systems that serve foundation models, covering throughput, latency, memory, reliability and cost. The person builds profiling and benchmarking tools, finds bottlenecks across serving, scheduling, distributed execution and accelerators, and works with other teams to land optimizations without hurting model quality. It suits engineers with hands-on production experience in ML serving stacks and accelerator-backed infrastructure.
- Function
- Development
- Estimated salary
- Est. ₹8–12 LPA
- Estimated for a mid-level role, ~3 yrs experience (typical for the level), India. Not published by the employer.
- Work mode and location
- On-Site, Bengaluru, India
Job description
Overview
We are looking for a Performance Engineer, Inference to understand and improve the systems that serve our foundation models. Inference is a tightly coupled system spanning model execution, serving runtimes, distributed systems, accelerators, scheduling, memory, and reliability. You will measure the system end to end, identify the highest-leverage performance gaps, and work across teams to close them while preserving correctness.
Roles and responsibilities
- Run cross-layer performance investigations across throughput, latency, memory efficiency, reliability, and cost.
- Build profiling, benchmarking, and observability tools that make inference performance measurable and explainable.
- Identify bottlenecks across model servers, batching and scheduling, distributed execution, memory systems, and accelerators.
- Partner with model, platform, and infrastructure teams to prioritize and land high-impact optimizations.
- Validate that performance improvements preserve model quality and numerical correctness.
Skills required
Top skills
- Performance profiling
- Python
- Distributed systems
- Multi-GPU inference
All skills (10)
- Transformer inference
- Accelerator performance
- Tensor parallelism
- Quantization
- KV-cache optimization
- vLLM
- SGLang
- TensorRT-LLM
- NVIDIA Dynamo
- Speculative decoding
Qualifications
- Hands-on experience profiling and optimizing ML systems or other performance-critical production systems.
- Production experience with at least one modern inference stack such as vLLM, SGLang,
TensorRT-LLM, NVIDIA Dynamo, or an equivalent serving runtime.
- Experience serving or operating large models across accelerator-backed infrastructure, including multi-GPU or multi-accelerator systems.
- Strong Python skills and the ability to read, instrument, and modify large production codebases.
- Solid understanding of transformer inference, distributed systems, latency/throughput trade-offs, and accelerator performance fundamentals.
Nice to have
- Experience with large-scale or multi-node inference, including tensor, pipeline, data, or expert parallelism.
- Experience with GPUs, TPUs, NPUs, or other ML accelerators and associated profiling tools.
- Experience with quantization, low-precision inference, KV-cache optimization, speculative decoding, or long-context serving.
- Experience contributing to or modifying inference runtimes, kernels, compilers, or distributed serving components.
- Experience optimizing inference for constrained or on-device environments.
Company info
About ZenteiQ.ai
ZenteiQ builds scientific foundation models grounded in physics, engineered for real-world industry. The sources do not name the customers or users of its models or products.