Goal

An LLM-based personal planning and scheduling assistant that can help users to manage their daily tasks and appointments. The assistant reads a default schedule for the week from weekly_schedule.md and can also be promtped to add, delete, or update tasks and appointments. The assistant can also be asked to provide a summary of the day’s tasks and appointments by brief me on my day command (with a date timestamp).

Design

The following data model are to be maintained by the assistant and optionally also the user:

• Default Weekly Schedule
  • The default weekly schedule is stored in weekly_schedule.md and is read by the assistant when needed.
• Temporary Schedule
  • The assistant also maintains a list of temporary, sudden, or one-time tasks added per user request. It will maintain them in temp_tasks.md and will read and write to this file as needed.
• User Notes
  • A list of user specified short text notes about tasks or user preferences that the assistant should remember. These notes are stored in user_notes.md and are read and written to as needed.

The following tasks are to be performed by the assistant, and the process is broken down into multiple agents to ensure correctness:

• Brief me on my day
  • The assistant should provide a summary of the day’s tasks and appointments on a given date.
  • Process: Interpreter identifies user request as this task, Readers reads data models and returns relevant information, Aggregator combines and format the information, and Briefer presents the information to the user.
• Remind me the rest of the day
  • The assistant should remind the user of the remaining tasks and appointments for the day at a given time.
  • Process: Interpreter identifies user request as this task, Readers reads data models and returns relevant information, Aggregator combines and format the information, and Briefer presents the information to the user.
• Update my schedule
  • The assistant should update the schedule with the user’s input: add, delete, or update items on the schedule.
  • Process: Interpreter identifies user request as this task, Readers reads data models and returns relevant information, Aggregator combines and format the information, Proposer proposes a list of changes to be made in the data models, Verifier checks the proposed changes for correctness (or provide feedback that loops back to Proposer; this loop will run for a maximum of 3 times), and Updater updates the data models, finally Notifier replies to the user with the changes.
• Suggest a time for a new task
  • The assistant should suggest a time for a task based on the user’s input. It either receives estimate of time needed or it estimates it directly, and it consults the schedule to find a suitable time slot.
  • Process: Interpreter identifies user request as this task, Readers reads data models and returns relevant information, Aggregator combines and format the information, Estimator estimates or proposes a different estimate than the user provided time estimate, Planner finds a list of suitable time slots for inserting the task, and Advisor presents the information to the user.