GuidePost (AI Coach for Workplace Conversations)

Problem and Motivation:

Workplace conversations are often a hidden source of conflict, inefficiency, and stress, yet most professionals receive little structured feedback on how they communicate. Misalignment, interruptions, and unbalanced participation can quietly degrade team performance and relationships, but these issues are difficult to quantify and improve without objective insight. This creates a clear need for a system that can both measure and guide communication quality in a practical, scalable way. The motivation behind this GuidePost (our AI coach) is to bridge that gap by combining conversational analytics with evidence-based guidance from cognitive behavioral therapy (CBT), enabling users to understand not just what went wrong in an interaction, but how to improve it. By translating complex communication dynamics into simple, actionable scores and personalized feedback, the product empowers individuals and teams to build healthier, more effective workplace interactions over time.

Data Science Approach:

RAG Approach:

Metrics:

Evaluation: 

We evaluated our system using anecdotal workplace interaction scenarios, including both strong and poor communication examples, to verify that the scoring metrics accurately reflected conversational quality and balance. We then assessed the AI-generated feedback through human evaluation, judging whether the advice was relevant, practical, and aligned with expected communication best practices. Overall, this combined approach ensured that both the quantitative metrics and qualitative guidance were meaningful, reliable, and useful to end users.

Key Learnings and Impact:

1.  Translating complex conversational behaviors into a simple scoring system made feedback more intuitive and actionable, helping users quickly understand and reflect on their communication patterns. 

2.  Integrating retrieval-augmented guidance grounded in CBT-based frameworks improved the relevance and practicality of feedback, enabling users to take clear steps toward behavior change. 

3.  The product showed strong potential to improve workplace dynamics by encouraging balanced, respectful, and goal-oriented conversations, ultimately supporting better collaboration and reduced conflict.

Acknowledgements:

We would like to thank all participants who contributed their time and feedback to test GuidePost and help improve its effectiveness.

• Advisor / Mentor Name : Fred Nugen & Ramesh Sarukkai
• Teaching Assistants :  Willam Li
• Additional acknowledgements : Michelle Hsu, Chris Lee, Kilian Komala, Jay Yee