Location:

Noida or Bengaluru

About the Role

This is the most technically differentiated research role in the company. You'll design the reward functions and RL algorithms that power the post-training behind the models we build — work that genuinely determines whether the models we ship are good.

What You'll Do
Design reward functions and RL recipes, including PPO/GRPO/DPO-family algorithms, for both our platform and our own model post-training
Ensure customer-supplied reward functions integrate cleanly and train stably on our platform
Diagnose and resolve RL training instability
Collaborate closely with our infrastructure team on rollout and setting requirements
Track recent RL research and make the call on what's actually worth adopting

What You'll Bring
8+ years of relevant experience, including PhD training, in Machine Learning,



Reinforcement Learning, or a closely related field — the underlying theory here takes years to build, and this is largely a PhD-level hiring bar
Deep, hands-on experience with modern RL algorithms (PPO, GRPO, DPO, or similar) and reward modeling
Proficiency in PyTorch or JAX
Robust statistical intuition for diagnosing why an RL training run is or isn't working

Education
PhD in Machine Learning, Reinforcement Learning, Computer Science, Statistics, Mathematics, or Physics, from a Tier-1 institution.

Nice to Have
A research background from a leading RL lab or applied AI company
Publication record in reinforcement learning or RLHF

📌 Research Scientist Rlhf/rl Bengaluru
🏢 HCLTech
📍 Bengaluru