LLM engineer

Pay in the destination countries

Monthly, in US dollars. Each country is an Imora admission package — open one for that route's fees, language requirement and deadline.

What does the day look like?

You work in a post-training team: you design and run fine-tuning and evaluation pipelines, move research methods such as RLHF into production, build measurement tooling, debug failed training runs, and write eval reports. On evaluations you build datasets, implement scoring, own dashboards, diagnose regressions during live runs, and harden distributed eval infrastructure.

Tools are Python, PyTorch, JAX, Transformers, DeepSpeed, vLLM, Triton, Kubernetes, Ray, Weights and Biases, MLflow, and cloud GPU clusters. Daily work is code review, experiment tracking, ablation analysis, data curation, cost monitoring per training run and per 1,000 tokens of inference. Prompting a clever model is under 10% of time, while around 90% is data cleaning, pipeline debugging, metric design, failed-run triage, and documentation. Postings list no regular night shifts, but a training regression needs response within hours, and on-call and production support during live training runs is explicit in Anthropic evaluations postings.

Where does this lead?