SR2
Founding Machine Learning Researcher (LLM Post-Training)
Founding Machine Learning Researcher (LLM Post-Training) at SR2, based in London. This is a permanent role with hybrid working.
- Pay
- Not given by the employer
- Hours and contract
- Permanent · Hybrid
- Location
- London
- Closing date
- Sector
- Research Scientist
Reference 642864209791821520
About the role
London - Hybrid
I'm hiring for an AI start-up focused on continual learning. They're looking for a Research Scientist/ Engineer to lead large-scale experimentation across supervised fine-tuning, reinforcement learning, distillation and repeated model adaptation.
You will take ownership of research questions, design training programmes and work directly with multi-billion-parameter language models.
What you’ll work on
- Train and post-train large language models.
- Design supervised fine-tuning and reinforcement-learning experiments.
- Research continual and sequential model adaptation.
- Build datasets, curricula and training-data mixtures.
- Evaluate capability improvements, retention and regressions.
- Work with distillation, synthetic data and model-generated training data.
- Develop RL environments and verifiable-reward tasks.
- Diagnose training failures and unexpected model behaviour.
- Build reliable experimentation and evaluation infrastructure.
- Collaborate with researchers developing new learning algorithms.
What we’re looking for
- Substantial hands-on experience training or post-training language models.
- Experience with SFT and reinforcement learning for LLMs.
- Knowledge of modern policy-optimisation methods.
- Experience with distributed or multi-GPU training.
- Strong Python and PyTorch, JAX or equivalent framework experience.
- Understanding of training dynamics, evaluation and generalisation.
- Ability to own research directions rather than only implement predefined approaches.
Particularly valuable experience
- Working with models in the 10B+ parameter range.
- Continual learning or repeated model adaptation.
- Distillation and synthetic-data generation.
- Reinforcement learning with verifiable rewards.
- Dataset construction and curriculum design.
- Building evaluation systems that detect capability loss and regression.
A PhD is valuable but not essential. Direct evidence of successfully training, adapting and evaluating large models is the primary requirement.
- Work on fundamental problems in language-model adaptation.
- Run ambitious experiments with direct influence over the research roadmap.
- Join a small, highly technical team at an early stage.
- Take meaningful ownership over both research and implementation.
- Receive a highly competitive salary and significant equity.
#J-18808-Ljbffr
Reference: 642864209791821520 · Posted 2 weeks ago · Closes 12 Dec 2026 · Listed via SR2
Apply for this job
This role is listed via SR2. Applications are handled on the employer's site.
Apply on employer siteOpens the employer's website in a new tab.
Signed in? Every job you apply for counts toward your weekly goal and badges. How it works
Safe applying: a genuine employer will never ask you to pay for a DBS check, training or equipment, or move you onto WhatsApp before you are hired. If this listing does, report it and do not pay anything.