Agentic AI for Energy Management – Research Consulting with Schneider Electric

Through the 2025 Applied Methods & Research Experience (AMRE) program, I collaborated with Schneider Electric's Data Science team, to develop and evaluate an agentic AI co-pilot for sustainable energy operations.

Advised by two faculty members, our core mission was to explore how large language model (LLM)-powered agents could serve as intelligent assistants for corporate energy management. We developed and tested multiple agentic architectures and benchmarked their performance on complex sustainability queries.

Our experiments demonstrated that single-agent systems often outperform multi-agent setups in execution time and accuracy for top-tier LLMs, while multi-agent systems provide scalability and modularity for mid-tier models and evolving toolchains. This nuanced finding underscored the importance of architectural choice in AI system design, a theme that resonates deeply with broader questions in agentic AI research.

This project stretched me across multiple dimensions:
<> I deepened my skills in LLM systems integration, multi-agent orchestration, and prompt engineering.
<> I gained hands-on experience with evaluation methodologies for intelligent systems, including both deterministic and LLM-based evaluators.
<> I learned how to balance technical experimentation with applied business needs, presenting findings directly to Schneider Electric Sustainability Business' data science team and refining solutions based on their feedback.