Inspiration
As a bachata dancer, I've spent a lot of time (and money) in lessons and wandering around social media wishing I could practice specific combinations at my own pace. Most dance students (me included) memorize a lot of moves but still have trouble chaining them together when they go for social dances. Professional choreography videos cost hundreds of dollars and aren't personalized to your skill level or music taste. You will not find a choreography that is just challenging enough for you to practice!
I thought: What if AI could be your personal dance instructor? What if you could just say "I want a romantic bachata routine with lots of turns for beginners" and get a custom video in seconds?
What if also, I could save my own library so I can remember my own moves?
Bachata Buddy is here to help students and trainers!
What it does
Creates personalized Bachata choreos based on a self authored 150 move library dataset
How we built it
Multimodal embedding concatenation to find the best matching video clips to the songs and user queries. Google AI to parse user queries into proper arguments I can use for the generation pipeline. Django Rest Framework Backend with React Frontend
Challenges we ran into
High CPU workloads was a problem: generating personalized dance videos with AI takes 3-5 minutes on CPU. That's too slow for a real-time learning experience. When I discovered the Cloud Run Hackathon and the GPU track was the perfect change to try new technology on my side project :)
Accomplishments that we're proud of
Deployment was more difficult than expected
What we learned
Multimodal embeddings, Frontend and Google Cloud technologies
What's next for Bachata buddy your AI dance teacher
- A knowledge graph of a larger library of moves
- A mobile app, I already have all of the necessary API endpoints
Log in or sign up for Devpost to join the conversation.