DDOI Search Engine

Inspiration

As an internet addict, I've found myself in conversations where I vaguely remember a clip from DDOI that would be relevant, but can't find it in time to share.

What it does

Using a vector database + an embedding model that scores over 64 on the MTEB test, we've compiled a list of clips and descriptions from Daily Dose of Internet. When you search a query, it puts it through the embedding model to find the most relevant clip(s).

How we built it

We cooked.

Challenges we ran into

For one, the internet was very shabby, and when doing testing I didn't even have to enable throttling in the Chrome devtools. Another thing was that we were only allocated about 8 hours to do the actual programming, which wasn't ideal.

Accomplishments that we're proud of

We are proud of actually finishing the project with an appealing UI, and for having the search engine return results that make sense.

What's next for DDOI Search Engine

We want to be able to import more clips easier, and to be able to have a clearer distinction between "relevant" and "not relevant clips" (currently, our vector database has a confidence of anywhere from 50%-80%, even for less relevant clips).

Built With

• firebase