Research engineer or scientist, post-training and reinforcement learning
You’d train and improve H’s language and vision-language models, including post-training for agents.
- London, about 3 office days a week
- Full time
- Post-training
- Pay not listed
This job is open.
Jack works for you and is free. Hiring companies pay us.
Jack’s take. Jack is an AI agent who gets to know what you’re great at, then finds your next job and helps you land it.
“I’d treat this as a post-training job for agents. It suits you if you’ve already trained a language or vision-language model.”
Who it suits
A good fit
- You have trained a language or vision-language model, with fine-tuning, preference optimization, or reinforcement learning.
- You can work from a training backend up to the learning method.
- You want the models used inside an agent rather than published as a result.
Think twice
- You haven’t trained a language model.
- You want an inference-only job.
- You don’t want about 3 days a week in the London office, or regular travel to another H Company office.
What you’d do
- You train language and vision-language models, including multimodal architectures.
- You implement training methods for instruction following and tool use.
- You build data and training systems for large distributed runs.
- You evaluate the models and work with other teams to put them into agents.
What they ask for
- You have trained a language or vision-language model, with fine-tuning, preference optimization, or reinforcement learning.
- You write Python, Rust, or a similar language, and can work from the training backend to the method.
- You use PyTorch, JAX, or TensorFlow.
How it works
- You’d be in the London office about 3 days a week, and travel to another H Company office about every 4 to 6 weeks. The posting doesn’t say which office.
What to know about this job
Open at least 15 weeks. Half of London AI engineer jobs have closed by week 9.
- Week 9Half of London AI engineer jobs have closed
- Week 15This job has been open this long
In November 2026, H Company had 2 research engineer jobs open whose titles name London.
The model builder, 23% of London AI engineer jobs
- 46%The product builder. Builds product features on large language models, from agents to search, and checks the answers hold up.
- 25%The platform builder. Builds the systems that train, serve and monitor models so they run fast and reliably at scale.
- 23%The model builder (this job). You train H Company’s language and vision-language models after pretraining, including instruction following and reinforcement learning, and put them into agents.
Shares of open AI engineer jobs in London.
Most new London AI engineers weren’t AI engineers before
- 71% of people who became AI engineers in London weren’t AI engineers in their last job.
- 47% of them came from a startup or scale-up.
That’s the wider AI engineer market. This posting is for someone who has already trained a language model.
H Company doesn’t list pay for this job
The London AI engineer guide shows what AI engineers in London expect to earn.
What London AI engineers earnWorth asking the hiring team
- “Which training methods are required on day one, and which can be learned here?”
Jack’s take on H Company
- It aims to work with 5 to 10 enterprise customers in 2026, up from 2 or 3 in 2025
- It raised a $220M seed round in 2024 from investors including Accel, Amazon and Samsung
- The team has more than doubled to over 100 people since 2024
- A former Palantir director runs it, with leaders hired from Palantir, Multiverse Computing and Meta
- Some of its people work 9 to 12 hours a day
- Based in Paris, London, New York and San Francisco
- Founded 2023
- 30 open jobs
Similar jobs in London
Other open jobs at in-demand startups.
About this page
Jack & Jill isn’t H Company’s recruiter. We summarized this job from H Company’s public careers page, where you can apply. We check whether it’s still open each time you open this page. The market numbers come from our London AI engineer guide and the company facts from H Company’s page. We last checked this summary against the posting on November 9, 2026. Jack’s take is Jack & Jill’s own.