LLM & Agentic AI Performance Engineer

Overview

Own end-to-end performance characterization and optimization of LLM and Agentic AI workloads, from Python-level frameworks and inference runtimes down to Linux, system architecture and hardware performance counters, within Intel's Data Center AI organization.

What you'll do

• Characterize and optimize LLM and Agentic AI workloads across the complete software and hardware stack.
• Identify performance bottlenecks across AI frameworks, inference runtimes, operating systems, memory, compute, and platform architecture.
• Design and develop hardware-aware and hardware-agnostic optimizations for AI frameworks and runtimes.
• Optimize LLM inference techniques including KV-cache management, attention, continuous batching, and quantization.
• Work hands-on with serving stacks such as vLLM, SGLang, TGI, or equivalent technologies.
• Analyze host and accelerator performance using tools such as VTune, EMON/PMU, perf, top-down analysis, and roofline modeling.
• Debug complex performance issues spanning Python frameworks, native code, Linux, runtime libraries, and hardware.
• Convert performance findings into software changes, upstream contributions, BKCs, and reusable optimization guidance.
• Collaborate with open-source communities to upstream and integrate performance improvements.
• Translate silicon-level findings into actionable recommendations for framework developers, and bring application and framework insights back to Intel platform and silicon teams.

What you'll need

• BE + 10 years, or M.Tech + 8 years, in Computer Science, Computer Engineering, Electrical Engineering, or a related discipline.
• Significant hands-on experience in performance engineering, systems optimization, or workload characterization on server-class platforms.
• Demonstrated ability to identify complex performance bottlenecks and deliver measurable performance improvements.
• Strong programming skills in Python and C++, or equivalent systems programming experience in C/Rust.
• Strong understanding of server architecture including cache hierarchy, memory bandwidth, NUMA, interconnects, and system-level performance behavior.
• Hands-on experience with Linux performance and system technologies such as affinity, huge pages, cgroups, eBPF, ftrace, perf, or equivalent tools.

Nice to have

• Deep understanding of LLM inference, including KV-cache, paged attention, continuous batching and quantization such as INT8, INT4, or FP8.
• Hands-on experience modifying or optimizing AI serving stacks such as vLLM, SGLang, TGI, OpenVINO, or equivalent.
• Experience building, profiling, or optimizing Agentic AI systems using technologies such as LangGraph, CrewAI, AutoGen, MCP-based frameworks, or similar platforms.
• Experience with hardware counter-level performance analysis using VTune, EMON/PMU, perf, top-down analysis, or roofline modeling.
• Contributions to significant open-source projects, including upstream code contributions.
• Ability to communicate across abstraction layers, translating silicon-level findings for software developers and application-level requirements for hardware architects.

Details

• Location: India, Bangalore.
• Work model: hybrid, splitting time between the assigned Intel site and off-site.
• Job type: Experienced Hire.
• Shift: Shift 1 (India).