No You've built human-data pipelines inside the machine. Come build the engine.
📍 Remote-friendly, Paris / London / USA focus · No defined budget + founding equity
🧠 RLHF · SFT · DPO · RL gyms · verifier & reward design · expert data ops
Most of the people who can do this job are doing it right now - inside a frontier lab or one of the big data vendors. They know how post-training data actually gets made, where it breaks, and why most of it isn't good enough.
This is the chance to stop running someone else's pipeline and own the whole thing.
We're an early-stage team building the training signal layer that frontier labs and enterprises can't build alone... grounded in real human behavioural data, not crowdsourced guesswork. The data that decides whether the next generation of agents actually works in the real world.
🚀 Founding-level ownership of the entire human-data function
🧬 Work shoulder-to-shoulder with researchers, not three layers away from them
💸 Unlimited salary cap + equity that actually means something (€30million valuation and have just closed significant funding)
🌍 Build for the labs defining the frontier
⚡ Genuine 0-to-1 - you're designing the system, not inheriting it
What you'll be doing
🔧 Standing up and scaling human-data pipelines for post-training - SFT, RLHF, preference data, RLAIF
👥 Recruiting, calibrating and retaining pools of high-skill domain experts - STEM, coding, finance, law
📐 Owning quality end to end - rubrics, gold sets, inter-annotator agreement, QA layers, verifier design
🤝 Translating model needs into data specs and collection protocols alongside research
🏗️ Building the taxonomies, task-authoring workflows and reviewer hierarchies the whole operation runs on
You'll love this if...
✅ You've built or run human-data ops for LLM post-training at a lab or major data vendor
✅ You've managed expert annotator pools, not just general crowdwork
✅ You think in rubrics, gold sets and agreement scores
✅ You've worked close to research and can speak both languages
✅ You want founding ownership, not another seat in a 200-person data org
This is one of those hires that gets made once. If it's you, you already know
.
Because great teams are Built Different.