Inspiration
Time management, no matter how good one is at it, contributes to mental load in one way or another. LLM Calendar Buddy is here to help! Users can just tell it all their tasks, and LLM Calendar Buddy will schedule everything for them. Now, you can focus on what, and not when.
What it does
The user will describe something they have to do, such as, "laundry in the evening" or "2 hour club meeting at 2 PM today." LLM Calendar Buddy will send this to Google Gemini, which has been trained to produce an "event object" containing data like the location, start and end times, and an event description. Then, this event object is passed to the Google Calendar API to actually be added to the user's Google Calendar.
How we built it
We went all in on the Google Cloud APIs, mainly using the Google Calendar and Gemini APIs. The project hinges on the "function-calling" capability that some more advanced LLMs like Gemini and GPT-3.5 have, where they can produce arguments for a user-defined function and even mock calling it.
Challenges we ran into
The biggest challenge, which still hasn't been entirely resolved, was getting Gemini to give us what we wanted. Despite all our attempts at "prompt engineering," it didn't always give us the desired "event object" we were trying to get from it.
Accomplishments that we're proud of
It's pretty awesome that we have a working and integrated front-end and back-end. Also, considering the whole LLM fervor going on since ChatGPT came out, we are proud of making an app that uses one.
What's next for Calendar Buddy
There is a lot of room for growth with this. The Gemini API has more functionality for pre-training and interacting with the model that we can use to produce more consistent output and more human-like decisions. We can also improve the user connection to Gemini through the app to have more communication, resulting in more detailed and accurate calendar event generation.
Log in or sign up for Devpost to join the conversation.