Built with LangGraph
This project implements an agentic AI workflow to automate email handling and responses using LangGraph.
The system intelligently reads incoming emails, classifies their intent, drafts responses, and sends them - with optional human-in-loop review for critical messages.
It utilizes GPT-OSS-120B as the underlying LLM which we deploy on the BHT compute cluster's Nvidia A100 80GB GPUs.
For a quick overview of the project (including photos of UI), look at Agentic_Email_Automation_Presentation.pdf
If you want to replicate the workflow (and have sufficient resources) examine the directories src/multi-agent-system and src/streamlit_ui as well as the .env.example and Makefile.
Note that the system primarily follows this tutorial with some modifications and plumbing / ops done to serve the LLM, create a front end, and wrap the agent in a service.