About Rapidata
Rapidata provides an API to humans that is revolutionizing the data generation and annotation industry. We deliver highly scalable, extremely fast human feedback that fuels the AI systems of the future — powering RLHF and DPO training data collection at internet speed for frontier AI labs. Our network reaches over 20 million active annotators across 192 countries, distributing micro-tasks ("Rapids") and returning verified labels in near real-time.
The Role
We're looking for a Reward Model Research Engineer to design, train, and productionize the reward models that turn Rapidata's real-time human feedback into high-quality training signal for RLHF and DPO. This is a research-to-production role: you'll work on the modeling problems that determine whether noisy, large-scale human preference data becomes a reliable reward signal — including process reward models (PRMs) that score multi-step reasoning rather than just final outcomes.
You'll be closing the loop between our human-in-the-loop data collection infrastructure and the reward models that consume it, tackling generalization across generator models, robustness to annotator noise, and inference-time guidance techniques like best-of-N and beam search — then validating your models in a live system operating at global scale.
What You'll Do
- Design, train, and evaluate reward models — including process reward models (PRMs) — from large-scale human preference data
- Build and improve inference-time guidance methods such as best-of-N sampling and beam search to make reward-guided generation more robust
- Develop training setups that improve generalization across diverse generator/policy models (e.g. multi-student style PRM training)
- Translate reward modeling research into production-ready pipelines that plug directly into Rapidata's RLHF/DPO data flywheel
- Collaborate with the data platform team to design data collection and active learning strategies that reduce reward model training bottlenecks
- Rigorously benchmark reward models for accuracy, robustness, and reliability before deployment
- Communicate findings clearly to both technical and non-technical stakeholders, including partner AI labs
What We're Looking For
- Hands-on research experience building reward models or process reward models (PRMs) for LLMs, ideally through an MSc/PhD thesis or equivalent applied research
- Solid understanding of reinforcement learning fundamentals and RLHF/DPO training pipelines
- Practical experience with inference-time guidance techniques (best-of-N, beam search) and evaluating multi-step reasoning
- Strong Python and deep learning framework skills (PyTorch)
- Experience turning research prototypes into validated, production-ready systems
- Solid statistical and mathematical foundation
- Excellent English communication skills, both oral and written
Nice-to-Have
- Experience with agentic system safety, guardrails, or LLM-based tool-calling agents
- Publications or open-source contributions in reward modeling, reasoning, or reinforcement learning
- Experience with hierarchical or model-based RL
- Based in or willing to relocate to Zürich
What We Offer
- Competitive salary and equity in a startup with strong growth, IP, and backing from top-tier VCs
- Opportunity to join a fast-growing startup early, with an outsized opportunity to shape where the company goes
- Opportunities for personal and professional growth as our team expands
- Fun and open (startup) culture
- Spacious mountain-view office located in Zürich Center near Sihlcity (3 min from Binz train station), with a large terrace, table tennis, pizza oven, hammock, and BBQ
- Hardware budget tailored to your preferences
- Unlimited snacks and drinks of your choice