NLP Performance Engineer

About G-Research

G-Research is dedicated to solving complex quantitative finance challenges by applying scientific rigor to financial intricacies. Based in our London headquarters, we assemble elite researchers and engineers who thrive in an environment that encourages thorough investigation and precise execution. We are committed to building a top-tier platform that amplifies our teams' innovative ideas.

Role Overview

As a member of our engineering team, you will help develop platforms and tools that support impactful research, enhancing scalability and innovation firm-wide. We are expanding our focus on large language models (LLMs) and NLP techniques, seeking a highly skilled NLP Performance Engineer to manage and optimize large-scale LLM inference performance.

This is a specialized quantitative development position where you collaborate closely with researchers to implement their ideas, with an emphasis on maximizing LLM performance. Your responsibilities will include designing and applying methods to boost the efficiency, cost-effectiveness, and capabilities of inference workloads on advanced computing hardware, enabling optimal usage of current and future infrastructures.

Key Responsibilities

• Analyze, benchmark, and tune large-scale LLM inference tasks on our compute resources
• Facilitate the efficient rollout of leading-edge models across diverse GPU platforms, adapting inference pipelines to evolving hardware
• Create and implement performance enhancement strategies for inference while preserving output accuracy
• Develop reference implementations, libraries, and tools to enhance the dependability and efficiency of NLP workloads
• Work collaboratively with researchers, leaders, and engineers to devise optimized solutions
• Partner with system, architecture, and platform teams to progress the compute stack and guide long-term platform strategies

Candidate Requirements

• Degree at Bachelor’s level or higher in Computer Science or equivalent practical experience
• Demonstrable expertise in profiling, benchmarking, and optimizing large-scale LLM inference workloads
• Methodical, evidence-based approach to performance tuning using strict benchmarking and reproducible methods
• In-depth knowledge of transformer inference mechanics, such as prefill versus decode operations, KV-cache functionality, attention variations, and performance constraints
• Practical experience with LLM serving frameworks (e.g., vLLM, SGLang, TensorRT-LLM, TGI) and proficiency with the PyTorch platform
• Familiarity with optimization strategies including quantisation, speculative decoding, and model parallelism on contemporary GPU systems
• Strong software development expertise, especially in Python and CUDA, building robust machine learning systems
• Excellent communication capabilities enabling effective collaboration across research, infrastructure, and engineering groups

Benefits and Perks

• Competitive salary package with an annual discretionary bonus
• Lunch provision through Just Eat for Business, plus access to a dedicated barista bar
• 35 days of annual paid leave
• Company pension contributions at 9%
• Casual dress code and supportive work-life balance
• Comprehensive healthcare coverage and life insurance
• Cycle-to-work program
• Regular monthly events for company-wide engagement