Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ScheduleMe: Multi-Agent Calendar Assistant

Oshadha Wijerathne, Amandi Nimasha, Dushan Fernando, Nisansa de Silva, Srinath Perera

TL;DR

ScheduleMe presents a graph-based, multi-agent calendar assistant that leverages LangGraph and LangChain to orchestrate specialized agents under a central supervisor, enabling natural-language calendar management via Google Calendar. The approach emphasizes modularity, context awareness, and scalability through distributed supervision and shared state, validated by zero-shot multilingual experiments and a user study. Key contributions include a detailed architecture, implementation stack, and empirical insights into multilingual performance, usability, and error modes, with practical implications for privacy, personalization, and real-world deployment. The work advances practical intelligent calendar tools by combining structured agent reasoning with conversational AI to improve usability, reliability, and flexibility in calendar management.

Abstract

Recent advancements in LLMs have contributed to the rise of advanced conversational assistants that can assist with user needs through natural language conversation. This paper presents a ScheduleMe, a multi-agent calendar assistant for users to manage google calendar events in natural language. The system uses a graph-structured coordination mechanism where a central supervisory agent supervises specialized task agents, allowing modularity, conflicts resolution, and context-aware interactions to resolve ambiguities and evaluate user commands. This approach sets an example of how structured reasoning and agent cooperation might convince operators to increase the usability and flexibility of personal calendar assistant tools.

ScheduleMe: Multi-Agent Calendar Assistant

TL;DR

ScheduleMe presents a graph-based, multi-agent calendar assistant that leverages LangGraph and LangChain to orchestrate specialized agents under a central supervisor, enabling natural-language calendar management via Google Calendar. The approach emphasizes modularity, context awareness, and scalability through distributed supervision and shared state, validated by zero-shot multilingual experiments and a user study. Key contributions include a detailed architecture, implementation stack, and empirical insights into multilingual performance, usability, and error modes, with practical implications for privacy, personalization, and real-world deployment. The work advances practical intelligent calendar tools by combining structured agent reasoning with conversational AI to improve usability, reliability, and flexibility in calendar management.

Abstract

Recent advancements in LLMs have contributed to the rise of advanced conversational assistants that can assist with user needs through natural language conversation. This paper presents a ScheduleMe, a multi-agent calendar assistant for users to manage google calendar events in natural language. The system uses a graph-structured coordination mechanism where a central supervisory agent supervises specialized task agents, allowing modularity, conflicts resolution, and context-aware interactions to resolve ambiguities and evaluate user commands. This approach sets an example of how structured reasoning and agent cooperation might convince operators to increase the usability and flexibility of personal calendar assistant tools.

Paper Structure

This paper contains 18 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Limitations of Traditional Scheduling Dialogue Systems
  4. Advancing to LLM-Based Multi-Agent Systems
  5. Multi-Agent Approaches to Scheduling Assistants
  6. Methodology
  7. Multi-Agent System Architecture
  8. Implementation Details
  9. Distributed Supervisor and Scalability Enhancements
  10. Experiments
  11. Results
  12. Zero-Shot Multilingual Evaluation
  13. User Study Metrics
  14. Qualitative Observations and Error Analysis
  15. Conclusion
  16. ...and 3 more sections

Figures (2)

  • Figure 1: Multi-Agent System Architecture: All other agents are controlled by the supervisor agent, but we have opted not to draw the control and communication lines between the agents to reduce unnecessary clutter. When a command or a data item is relevant to all the entities in a parent entity, the relevant arrow terminates on the parent entity. Otherwise, it terminates on the specific relevant child entity. The numbers on the agents at arrow terminals indicate the order in which each action may happen in a typical execution.
  • Figure 2: AI Calendar Assistant Interaction Example: A representative dialogue flow demonstrating the assistant's ability to process user queries related to calendar management. The supervisor agent interprets the user’s request and delegates actions to appropriate sub-agents (e.g., availability checking and event modification). The updated calendar view on the right confirms the successful execution of the rescheduling task.