AI Agent

ITAI 2376: Deep Learning
Final Report: Virtual Administrative Assistant AI Agent
Team Name: Epoch Explorers

Project Overview

Our capstone project for ITAI 2376 was to design and implement an AI agent system that could automate multistep administrative processes using language-based reasoning and decision-making. We chose Process Automation Agent and created a Virtual Administrative Assistant that performs tasks like scheduling meetings and sending emails in response to natural language user input. Output creation and logging are done using OpenAI's GPT model for reasoning and planning, n8n for process execution, and external applications including Google Calendar, Gmail, and Google Sheets. The agent also includes built-in error handling and reinforcement learning elements via outcome tracking. This project let our team work together in a cloud-based development environment applying course ideas in a practical automation context.

Agent Architecture

Our agent follows a modular structure with the following components:

Input Processing: The agent receives natural language messages via a n8n webhook trigger.
Reasoning Component: OpenAI's GPT model is used to determine the user's intent and decompose high-level instructions (e.g., "schedule a meeting") into an action, message, date, and time.
Conditional Logic: A combination of If and Switch nodes in n8n determine whether to create a calendar event, send an email, or log an unrecognized action.
Output Execution: The agent takes action by either creating a Google Calendar event or sending an email via Gmail.
Confirmation: After completing a task, a confirmation message is sent back to the user.

Tool Integration

We successfully integrated several key tools to build our agent:

Google Calendar API: Used to create meeting events based on the given date, time, and user details.
Gmail API: Used to send emails with dynamically generated content.
Google Sheets: Logs unrecognized or unhandled requests with details like the action, message, date, and time for future review.
n8n Automation Platform: Served as the core workflow engine to orchestrate the various components and API integrations.
OpenAI GPT: Provided the natural language understanding and reasoning capabilities to interpret user input and generate appropriate actions.

Development Environment

We began with custom NLP tools and Python-based logic to interpret language input and automate workflows like e-mail responses, task creation, and calendar scheduling. The project later transitioned to the n8n automation platform, enabling us to visually design, test, and deploy the workflow without requiring a local server setup. This shift allowed faster development and integration of modular workflows without rebuilding core logic from scratch.

Course & Team

Project Overview

Agent Architecture

Tool Integration

Development Environment