Inspiration
In a world where AI is taking the lead, we began exploring the many applications of a digital twin, and how they can guide choices before they even happen.
What it does
We create a digital twin of you from a few simple onboarding questions. Your twin interacts with other digital twins of real people, simulating events, conversations, and even the people you might meet. It then predicts how much you’ll enjoy the experience and gives a clear recommendation: go or skip.
How we built it
Our main bread and butter is our multiagent simulation network using Claude. Up to 7 agents representing people of different personalities, interests and backgrounds all come together, and are learnt and simulated together to form emerging lifelike interactions. Our front end was made with react, and we used APIs (FourSquare, Eventbrite) and llms to collect metadata for events, and restaurants. We used sqlite3 with vector embeddings to store people and events.
Challenges we ran into
- Having to develop with three different OSes. Some tech like PostgreSQL was unfortunately very hard to get working on all of our laptops.
Accomplishments that we're proud of
Created a platform which genuinely predicted cool events which we would attend.
What we learned
How to develop with LLMs, databases, vector embeddings etc.
What's next for EventTwin
Hopefully scale our product to contain more twins, more events, and more locations. The Digital twin idea can be used not just for event prediction, but can also be used in management, dating etc. There are a lot of very interesting applications
Built With
- claude-hussenet-anagram
- langchain
- langgraph
- masa
- python
- react
- sqlite
Log in or sign up for Devpost to join the conversation.