TA Copilot

Inspiration

As educators face increasing email volumes from students, we saw an opportunity to leverage AI to help teaching staff manage their inbox more efficiently while maintaining personalized, thoughtful responses.

What it does

AI Teaching Support Agent analyzes student emails and provides three key capabilities:

Summarization: Condenses lengthy emails into key points
Classification: Categorizes emails by type (question, complaint, request) with routing recommendations
Reply Drafting: Generates professional, context-aware response suggestions

How we built it

Frontend: Streamlit for rapid prototyping and clean UI
AI Engine: Amazon Bedrock Converse API with multiple model options (Claude 3.5 Sonnet, Haiku, Amazon Nova)
Infrastructure: AWS credentials integration for seamless Bedrock access
Design: Structured JSON output for classification enables future automation workflows

Challenges we faced

Balancing AI creativity with professional tone requirements for educational settings
Designing prompts that work consistently across different Bedrock models
Structuring classification output to be both human-readable and machine-parsable

What we learned

Amazon Bedrock's Converse API provides excellent flexibility for multi-model experimentation
Streamlit enables incredibly fast iteration for AI-powered prototypes
Proper prompt engineering is critical for consistent, reliable AI responses