ML Systems Engineer – Model Training and Infrastructure (SWE-focused LLMs)
The role is for an engineer who can write solid production code in Python and Go and also work close to model training in PyTorch, TensorFlow or JAX. Day to day it covers building fine-tuning and reinforcement learning pipelines, creating training data, running containerised services, and checking whether new models really get better at software tasks. A strong candidate has roughly three to five years of experience, cares about code quality as much as model scores, and can explain experiment results clearly.
How it fits the AI era
Here AI is both the product and the thing being built, so the work is about shaping how coding models learn rather than just using them. Much of the effort goes into data, test environments and reward design, where small choices change how a model behaves. Human judgement stays central in deciding what good, maintainable code looks like, spotting when a model games its reward, and turning failures into better experiments.
How it builds AI experience
Someone in this role would gain hands-on experience training and evaluating coding models, from fine-tuning to reinforcement learning with real repositories and test runs. Because the company builds for locked-down settings such as banks and defence, there is also a chance to learn how to post-train open-source models that must run privately. To build this beforehand, try fine-tuning a small open model on code tasks and building a simple test-based reward loop.
Worth showing
- Python and Go production code
- Custom PyTorch training loops
- Docker and Kubernetes services
- Clear experiment write-ups
This is our own read, written from the employer’s posting and the company’s Spotlight. The employer’s posting is the source of truth for the role.
About Cosine
An AI coding agent built to run entirely inside a company’s own secure or air-gapped environment, aimed at banks, defence contractors and anyone who cannot send code to a public cloud model.
#J-18808-Ljbffr…