Training Smarter Agents: Evaluation, Feedback Loops, and RL

Hack Session

About the session

Most agentic systems rely on static prompting, yet real-world environments require agents that can adapt and improve over time. This workshop presents a practical approach to training smarter agents using evaluation-driven feedback loops and lightweight reinforcement learning techniques.
 
Participants will build a baseline agent, implement behavior-level evaluation, and introduce feedback mechanisms to iteratively improve performance. We will cover trajectory comparison, bandit-style optimization, and system-level reinforcement learning methods that enhance decision-making without retraining large models.
 
By the end of the talk, attendees will have a working framework for building agents that not only perform tasks but continuously improve in production environments.

Speaker

Download Brochure