Research Scientist - RLHF, RLAIF & Reward Modeling

This role is for one of Weekday’s clients
Salary range: Rs 5000000 - Rs 10000000 (ie INR 50 - 100 LPA)

Min Experience: 3+ years
Location: Bengaluru, Karnataka, India
JobType: full-time

We are looking for a highly skilled and research-oriented Research Scientist with 3–6 years of experience in machine learning, reinforcement learning, and large language model (LLM) alignment. The ideal candidate will have strong hands-on experience with Reinforcement Learning from Human Feedback (RLHF), Reinforcement Learning from AI Feedback (RLAIF), and Reward Modeling, and will contribute to developing and improving advanced AI systems.

You will work on research problems related to model alignment, preference learning, reward optimization, evaluation, and post-training. This role requires a strong understanding of modern machine learning techniques, the ability to translate research ideas into working systems, and experience conducting rigorous experiments on large-scale models.

Requirements

Key Responsibilities

• Design, implement, and evaluate RLHF pipelines for training and aligning large language models with human preferences.
• Develop and improve RLAIF methodologies using AI-generated feedback, preference signals, and automated evaluation frameworks.
• Build, train, and validate reward models that accurately capture human or AI preferences and desired model behaviors.
• Experiment with reinforcement learning and preference optimization techniques to improve model helpfulness, accuracy, safety, and instruction following.
• Analyze model behavior and training outcomes using quantitative evaluations, benchmarks, and controlled experiments.
• Develop data-generation, preference-collection, ranking, and annotation strategies for alignment and post-training datasets.
• Collaborate with research engineers and ML engineers to scale training and experimentation pipelines.
• Investigate failure modes in reward models, preference datasets, and alignment techniques, and propose research-driven solutions.
• Stay current with emerging research in LLM alignment, reinforcement learning, preference learning, reward modeling, and AI feedback.
• Document experimental results and communicate research findings clearly through technical reports, presentations, and research papers.

Must-Have Skills

• 3–6 years of hands-on experience in machine learning, deep learning, reinforcement learning, or a closely related research field.
• Strong practical experience with RLHF (Reinforcement Learning from Human Feedback).
• Strong understanding and hands-on experience with RLAIF (Reinforcement Learning from AI Feedback).
• Proven experience developing, training, or evaluating reward models and preference-based learning systems.
• Strong understanding of reinforcement learning concepts, policy optimization, reward functions, preference modeling, and model evaluation.
• Experience working with Large Language Models (LLMs) and their training or post-training workflows.
• Strong Python programming skills and experience with modern deep learning frameworks such as PyTorch or equivalent.
• Ability to design experiments, interpret results, troubleshoot training issues, and derive meaningful research insights.
• Strong mathematical and statistical foundations relevant to machine learning and reinforcement learning.

Good-to-Have Skills

• Experience with PPO, DPO, GRPO, or other reinforcement learning and preference optimization techniques.
• Experience working with transformer architectures and LLM fine-tuning.
• Familiarity with distributed model training and large-scale experimentation.
• Experience publishing research papers or contributing to open-source ML research.
• Knowledge of model evaluation, red-teaming, AI safety, or alignment research.

Qualifications

A Master’s or Ph.D. in Computer Science, Artificial Intelligence, Machine Learning, Mathematics, Statistics, or a related technical field is preferred. Candidates with strong industry research experience and demonstrated expertise in RLHF, RLAIF, and reward modeling are encouraged to apply.