tudum

inspiration

i binge a lot of movies, tv shows. like a lot. often i come across scenes on ig or tiktok, when i wanna know which movie it’s from. some drop movie names in comments. but that's not always the case :(

what it does

it identifies movies based on a short sample played using the microphone+screen capture. on mobile, you can access the app from the notification bar over any app (ex - instagram) and it'll detect the movie, tv show while you're watching reels.

how i built it

-> kotlin (android app) -> python (backend) -> supabase (postgresql db) -> ffmpeg (video pre-processing)

challenges i ran into

-> latency issue: the first test took 20secs to detect. then i implemented real-time processing to bring that to 2-3secs. -> video pre-processing: so far i had to deal with 10k movies to train the ai. the bottleneck was speed i.e. bandwidth, compute. i had to optimize at each level to make it feasible and fast.

what i learned

i got hands-on building android app in kotlin and the entire ai architecture. i'm grateful :)

what's next for tudum

tbh idk :)

Built With

• fastapi
• ffmpeg
• hugging-face
• machine-learning
• openai