Lead LLM Inference Engineer - Remote/UAE

HellaGoodAudio; United States; Remote; USD 180,000 - 230,000; Full time

Benefits offered by this job

Technical ownership

Competitive salary

24 days annual leave

Health insurance

Modern office in Yas Creative Hub

Job summary

Dizzaract, a product-driven company at the intersection of gaming, digital platforms, and AI, is seeking a Lead Inference Engineer for FAR Labs. You will own the technical direction of the inference stack, optimize performance, and drive measurement-driven improvements across serving, GPU, and distributed systems.

This senior role requires hands-on leadership, strong experience with LLM inference serving, CUDA/Triton, MoE, quantization, and benchmarking.

Qualifications

• Hands-on experience building or operating production LLM inference-serving systems.

Responsibilities

• Own and improve the FAR Labs inference serving stack and related optimisations.
• Improve throughput, latency, memory efficiency, and cost while upholding model quality.
• Identify GPU bottlenecks and optimise via CUDA, Triton, profiling, and memory strategies.
• Drive efficient MoE serving including expert parallelism and load balancing.
• Define measurement methodologies and establish reproducible benchmarks.
• Lead engineering standards for inference performance and production readiness.
• Mentor engineers across serving, metrics, and testing.

Skills

LLM inference serving

GPU performance engineering

Distributed systems

Benchmarking

Technical leadership

Model quantization

CUDA/Triton

MoE inference

Tools

CUDA

Triton

TensorRT-LLM

vLLM

SGLang

KV-cache

Job description

Dizzaract, a product-driven company at the intersection of gaming, digital platforms, and AI, is seeking a Lead Inference Engineer for FAR Labs. You will own the technical direction of the inference stack, optimize performance, and drive measurement-driven improvements across serving, GPU, and distributed systems.

This senior role requires hands-on leadership, strong experience with LLM inference serving, CUDA/Triton, MoE, quantization, and benchmarking.