Machine Learning Engineer at Cosine.sh

Jack & Jill/External ATS; Greater London; On-site; GBP 70,000 - 90,000; Full time

Job summary

A pioneering AI software development company in London seeks a Machine Learning Engineer to lead the training of advanced software engineering models. The candidate will design large-scale training experiments using PyTorch and multi-node GPU clusters, contributing to the evolution of intelligent coding capabilities. Essential requirements include 3–5+ years of experience in deep learning, especially with LLMs, and a strong foundation in distributed training. This role offers substantial opportunities to influence AI-driven software engineering tools.

Qualifications

• 3–5+ years of experience training deep learning models in production environments.
• Deep proficiency with PyTorch and experience with torch.distributed.
• Experience training large sequence models or LLMs with scaling understanding.

Responsibilities

• Transform open-source models into high-performance Lumen Enterprise SWE agents.
• Design and execute large-scale training experiments on multi-node GPU clusters.
• Build and refine large-scale RL loops for code generation and performance improvement.

Skills

Training deep learning models in production environments

Proficiency with PyTorch

Experience with multi-GPU and multi-node training

Training sequence models or LLMs

Job description

This is a job that we are recruiting for on behalf of one of our customers.

To apply, speak to Jack. He's an AI agent that sends you unmissable jobs and then helps you ace the interview. He'll make sure you are considered for this role, and help you find others if you ask.

Machine Learning Engineer

Company Description: Cosine.sh is pioneering AI for software engineering, building advanced open-source-based models that empower developers with intelligent, real-world coding capabilities.

Job Description: Join Cosine.sh as a Machine Learning Engineer and lead large-scale training of Lumen Enterprise models—our flagship software engineering LLMs. You will drive state-of-the-art performance through supervised fine-tuning and reinforcement learning, operating close to the metal with PyTorch, distributed systems, and long-context architectures. This is a high-impact role where your work will directly shape the capabilities engineers depend on daily.

Location: London, UK

Why this role is remarkable:

• Directly influence the next generation of Lumen Enterprise SWE models used by engineers every day.
• Operate at real scale: modern open-source models, long-context training, MoE architectures, and multi-node GPU clusters.
• A true full-stack ML engineering role combining PyTorch, distributed systems, data pipelines, RL design, and MLOps.

What you will do:

• Transform open-source base models into high-performance Lumen Enterprise SWE agents using supervised fine-tuning and RL.
• Design and execute large-scale training experiments on multi-node GPU clusters, including long-context and MoE training.
• Build and refine large-scale RL loops where models write code, run tools/tests, and receive reward signals to improve performance.

The ideal candidate:

• 3–5+ years of experience training deep learning models in production environments.
• Deep proficiency with PyTorch and strong hands-on experience with torch.distributed for multi-GPU and multi-node training.
• Experience training large sequence models or LLMs (70B+ parameters), with practical understanding of scaling challenges.

How to Apply:

To apply for this job speak to Jack, our AI recruiter.

1. Visit our website
2. Click "Speak with Jack"
3. Login with your LinkedIn profile
4. Talk to Jack for 20 minutes so he can understand your experience and ambitions
5. If the hiring manager would like to meet you, Jack will make the introduction.