ClassPart AI
AI agents that listen to a class discussion and draft evidence-backed participation feedback for the professor to approve.
Lead developer · SMU IS469 Generative AI with LLMs
What it is
A seven-step multi-agent workflow: a listener turns the recording into speaker-labelled turns, a coordinator plans the work, a research agent runs agentic retrieval over course material, and a judge, an auditor and a feedback agent turn the evidence into feedback. Nothing reaches students until the professor approves it.
What I did
Led development and authored most of the merged pull requests. I deployed the demo on my own server and kept it running through the team’s changes.
What I learned
Agents need budgets and guardrails as much as prompts. Every run has a hard cap on model calls, tools and time.
Built with
- Python
- OpenAI
- Streamlit
- faster-whisper
- Docker