Home
FUTURE STUDENTS ask.UWindsor GIVE TO UWINDSOR
Faculty of Science

Agentic AI Workflows: Reflection Design Pattern (1st Offering) - JLR Challenge #4 Technical Workshop

Tuesday, November 11, 2025 - 15:00
School of Computer Science – JLR Challenge #4 Technical Workshop

 

Agentic AI Workflows: Reflection Design Pattern (1st Offering)

Presenter: Mehedi Hasan Shanto

Date: Tuesday, November 11th, 2025

Time: 3.00 PM

Location: Workshop Space, 4th Floor - 300 Ouellette Ave., School of Computer Science Advanced Computing Hub

 

Abstract: 
This workshop, part of the JLR Hackathon Mentorship Series, introduces students to the Reflection Design Pattern in Agentic AI workflows.  Participants will explore how reflection—where a model evaluates, critiques, and iteratively improves its own outputs—can significantly enhance performance in real-world tasks such as code generation, data visualization, and prompt engineering.

The session builds on the foundational concepts (Introduction to Agentic Workflows), transitioning from basic task decomposition and planning to autonomous self-improvement mechanisms. Through live demonstrations and hands-on exercises, attendees will learn to design reflection prompts, interpret iterative refinements, and evaluate their models’ improvements using rubric-based and code-level evaluation methods.

Workshop Outline:
  •  Introduction to Agentic Workflows
  •  Review of task decomposition, tool use, and planning design patterns.
  •  Understanding Reflection Design Pattern
  •  Concept: self-improvement loop inspired by human reflection.
  •  Real-world analogy: refining an email or code through feedback.
  •  Reflection vs. Direct Generation
  •  Comparison of workflows using coding and text generation examples.
  • Demonstration: Reflection for Code Improvement
  •  Example: LLM writes initial code, identifies bugs, and revises the second draft.
  •  Hands-On Mini Exercise
  •  Participants modify a reflection prompt to improve LLM output quality.


 

Prerequisites:


• Basic familiarity with Python and prompt-based AI systems.
• Access to Google Colab or Jupyter Notebook for code demonstration
 

Biography: 

Mehedi Hasan Shanto is a Ph.D. student in the School of Computer Science at the University of Windsor, specializing in software engineering, large language models (LLMs), and repository mining. His research focuses on understanding and automating software evolution by combining empirical software analysis with AI-driven techniques. He has extensive experience mining large-scale GitHub datasets and building LLM-based frameworks for edit prediction and code quality evaluation. Shanto’s work aims to make software repositories not just storage systems, but dynamic datasets that reveal insights about development processes and developer behaviour.

 

Registration Link (Only MAC students pre-register)