Idetic

thumbnail
video library page
video search page

Inspiration

Remember that moment from your sibling’s birthday party you wanted to share with your family? Now imagine sorting through hundreds of birthday videos and hours of footage to find it. That’s what inspired us to create Idetic.

What it does

Idetic is a video search engine. Upload your video to our video library, and we automatically categorize it into relevant chunks, allowing you to search through your videos as if they were text.

How we built it

We built Idetic through the use of a Convex & Next.js backend with integrations with Auth0 for authentication, and Convex's database for file storage. Our ML components use models like Whisper and clip running on a machine, layered with Convex and FastAPI for full-stack communication. Audio processing was handled with ffmpeg, and video processing with OpenCV2. We then performed vector similarity searches to the generated embeddings through Convex offerings.

Challenges we ran into

We ran into major challenges with creating vector embeddings on the backend since video based vector storage can be a lot more computationally expensive.
Hosting on Convex and integrating file storage was harder than we anticipated. We also struggled with querying the database as it uses a system quite different to what we worked with
We also had difficulty evaluating performance with different param

Accomplishments that we're proud of

We're super proud of the fact we managed to string together a super complex project as far as we did within the 36 hours given. Vector embeddings for videos can get quite difficult and building recommendation systems alongside search engines is a challenge.