Skip to content
Search jobs inside your AI agent. Claude, ChatGPT, Claude Code, Codex, Gemini, Cursor and Grok
Connect

Built Different

NewBe an early applicant

Founding Research Engineer (RL)

Posted

Description

Founding RL Engineer: Post-Training the Agents Frontier Labs Learn From

📍Paris, hybrid working

💰 Up to €170,000 + founding equity

🧠 Open-weight post-training (SFT, DPO, GRPO) · agentic models · tool use · reward design


Plenty of people have fine-tuned a model. Very few have taken an open-weight model, post-trained it into an agent that does real work, and shipped it somewhere it actually mattered.

If that’s you, keep reading.


Anthropic leadership has discussed spending over $1 billion on RL environments in a single year.


The market is moving fast. One pre-seed company (my client today) is about to close a round at a valuation most Series A companies would envy, with frontier AI labs already paying and acquisition interest on the table twice over.


The reason? Most teams build training environments from synthetic data. Easy to demo, easy to copy, brittle. This team captures how real domain experts reason, decide and solve long, complex tasks, and turns it into the environments, rewards and evals the best labs learn from. 10 to 100+ step workflows. Closed loops where data, training and evaluation are tightly integrated.


The results: 25 to 30% uplift in task success, 50 to 65% more training signal, and evals that reflect how people actually work.


Why this one


🚀 €170,000 + founding equity at a ~$40M seed valuation

🧠 Frontier AI labs as paying customers from day one

🤖 You post-train the agents yourself, you don’t just prompt them

🏗️ Environments, rewards and evals treated as first-class problems

📈 Ground floor, before the rest of the market catches up


What you’ll be doing


🔧 Fine-tuning open-weight models (Qwen, Llama, Mistral and others) with SFT, DPO and GRPO

🤖 Building agentic models that use tools, plan and act across long multi-step tasks

🎯 Designing rewards and verifiers from real expert workflows

🧪 Building the evals that show where agents break, then proving the fix

🚢 Taking it from experiment to something a frontier lab or enterprise actually runs


You’ll love this if...


✅ You’ve fine-tuned open-weight models yourself, not just called a closed API

✅ You’ve built agentic models: post-training agents to use tools and finish multi-step tasks, not wiring up RAG

✅ You’ve done it at an enterprise, a big-name tech company or an AI lab, where the scale and stakes were real

✅ Two of those three gets us talking. All three, and we should speak today

✅ Your RL is the genuinely hard kind: reward design, credit assignment, long horizons

✅ You move fast, think from first principles and want to build at the frontier


We are hiring from Junior to Senior! Apply today!

Because great teams are Built Different

Skills

Apply

See also