Its main product is a large AI companionship platform. An internal AI moderation system, built in-house, governs what the company builds and checks that it is safe and ethical.
The founders have built and scaled web products from zero to IPO. Leadership brings 10+ years of technology executive experience, 8+ years in customer acquisition and e-commerce growth, and 25+ years in finance, strategy and M&A. The CTO has 3 years of work in AI and LLMs.
The LLM team is small and led by Marco, the LLM Tech Lead. It works with the CTO and the product, web and DevOps teams, across data, post-training, evals and model behavior, from failure analysis through to production.
Most traffic runs on self-hosted, fine-tuned open-weight models. On the company's internal roleplay evals, smaller fine-tuned models have beaten much larger off-the-shelf models.
Responsibilities
Fine-tune open-weight LLMs with SFT, LoRA or QLoRA, and post-training methods such as DPO and GRPO.
Work out why a model fails and design experiments to fix it, deciding whether the answer lies in data, post-training, prompting or classifiers.
Build training data and evals that feed directly into what gets trained next.
Improve how models handle personalization, character consistency, tone, multilingual conversation and group interactions.
Ship models to production and use real user behaviour to shape the next iteration.
Shorten the training loop so experiments run faster, are easier to compare and can be reproduced.
Requirements
You have post-trained open-weight LLMs for a user-facing product and shipped them to production.
You have worked somewhere AI was central to the product or business, not a side experiment.
Fine-tuning is a core part of your experience.
You are strong in Python, PyTorch and Hugging Face, and have hands-on experience training and debugging on GPUs.
You build your own data and evals and make decisions based on evidence.
You can move work out of notebooks and into production.
You use AI tools every day in your own engineering work.
You are fully comfortable working with uncensored models and explicit NSFW content, including harmful material, when developing and testing moderation systems.
Bonus: experience with conversational AI, roleplay products or highly personalized consumer AI.
Bonus: experience with multilingual LLMs.
Bonus: experience in moderation, safety or adversarial testing.
Bonus: experience with multimodal models.
Bonus: contributions to open-source LLM or post-training work.
Conditions
The role is 100% remote, and you choose the environment where you work.
The company prefers B2B contracts but is flexible. What matters most is a long-term commitment and impact.
Benefits
20 working days of paid time off per year.
A monthly allowance of 100 USD towards health insurance costs.
Unlimited 1:1 sessions with psychologists and lifestyle experts through a third-party wellbeing provider, also open to up to three family members.
A co-working budget of up to two visits per month, at 35 EUR or 40 USD per visit.
A learning budget for courses, books, conferences, events or certifications.
A company laptop, plus a monitor budget of up to 250 USD and a budget of up to 150 USD for keyboard, mouse, mic and headphones.
Premium access to ChatGPT, Hugging Face, Claude Code and any other tool you need for your work.
A yearly in-person get-together for the team to connect, brainstorm and celebrate wins.
Where you’d work
Fully remote
You can work from
United Kingdom
About the company
Hidden companyCompany hidden
Industry: AI
Good
Your chances
Still hiring, not crowded yet, and you'd be among the first.
17 checks run
5 good signs
0 red flags
Still hiring?
11 checks
Actively hiring
StaleHiring
In its favour4
Still on the company's own careers site, checked 1 h agoModerate evidence
Found in the last 48 hours, before the big job boardsModerate evidence
Specific about the basics: pay, place, level, stack and contract all statedSlight evidence
1 moreFewer
States its salarySlight evidence
How crowded?
6 checks
Low
QuietCrowded
In its favour1
In its Early Window: not on the big job boards yetStrong evidence
?
Fits Me
How well does this role fit you?
Answer a few questions or drop your CV, and every role gets a fit score with the reasons, this one first.