Post-Training and Reinforcement Learning Engineer

About the Role

SustainabilityEconomics.ai is seeking a Post-Training and Reinforcement Learning Engineer for its platform team in Bengaluru. This early-career position involves training, evaluating, and improving open models for Clem.ai, an Energy-Mix Agent. The role reports to the Post-Training Lead and involves collaboration with senior engineers and energy experts.

Key Responsibilities

• Prepare, clean, and verify training data for tasks such as bill and invoice extraction, contract clause tagging, and optimization template completion.
• Run supervised fine-tuning of open models using LoRA on company GPU infrastructure.
• Develop reinforcement learning environments and reward functions.
• Evaluate models before and after training on the company’s energy-domain benchmark and report results.
• Convert operator corrections into new training examples and evaluation cases.
• Generate and validate synthetic data.
• Support the distillation of larger models into smaller ones.
• Maintain reproducible experiments, including configurations, data versions, run logs, and summaries.
• Use coding agents for code, tests, and data preparation, reviewing all output.
• Comply with data governance requirements.

Requirements & Tech Stack

• Strong proficiency in Python.
• Working knowledge of PyTorch.
• Demonstrated experience fine-tuning at least one open language model (e.g., with Hugging Face Transformers and LoRA).
• Sound understanding of machine learning fundamentals (data splits, overfitting, evaluation metrics).
• Working knowledge of reinforcement learning fundamentals (policies, rewards, environments).
• Proficiency with Git, Linux, and GPU-based training jobs.
• Rigor in data handling and evaluation.
• Ability to report results objectively.
• Clear written communication in English.

Qualifications

• Bachelor’s or master’s degree in computer science, electrical engineering, mathematics, or a related discipline.
• Up to two years of relevant experience, including internships and research projects.
• Prior experience in the energy sector is not required.

Preferred Background

• Practical experience with preference-based or reinforcement learning methods for language models (e.g., DPO or GRPO-style training).
• Familiarity with reinforcement learning algorithms like PPO or SAC, and tools such as Gymnasium.
• Experience writing reward functions or automated graders.
• Experience with distillation or synthetic data generation.
• Experience with structured outputs and tool calling in language models.
• Exposure to optimization solvers or time-series forecasting.
• Interest in power markets, renewable energy, or data center operations.
• Open-source contributions, competition results, or published research.

Personal Attributes

• Leadership skills and ability to take ownership.
• Strong self-motivation and initiative.
• Eagerness to learn new methods and a new domain quickly.
• Effective teamwork.
• Accountability for results and integrity in reporting.
• Ability to manage time and deliver to deadlines.

Location

• Bengaluru