WANsearch
Tharusha Jayasooriya
A search engine for the WAN show podcast. Search for a phrase / words said in the show and get time stamps to the exact moment. It uses techniques such as Inverted indexing and Cosine-similarity to find videos.
Frontend
The frontend is written in svelte. And I recently rewrote this frontend in React as well.
Backend
Written in Golang using Gin.
Database
Using a SQLite database and the driver by mattn.
How it works
The algorithm currently uses these techniques to rank documents. (in this order)
Inverted Index
Cosine similarity
Word proximity
Title weight (If the title includes a query term it is rated higher)
When retrieving relevent documents,
And uses a simple LIKE sql statement in all_dialogues table for quoted search.
Infrastucture
Frontend is hosted in Cloudflare pages.
The API is running in EC2. Uses Cloudflare proxy.
Like this project
Posted Aug 20, 2024
A custom search engine for a podcast listeners.
Likes
0
Views
8
