I Shipped a Content-Based Fanfiction Recommender Instead of My Planned Taste Engine

Short summary

The author built Opsis, a scoped-down content-based recommender for fanfiction that finds similar works using embedding cosine similarity blended with fandom and relationship overlap. It scrapes AO3 metadata, stores 7,000+ fics in PostgreSQL, embeds summaries with all-MiniLM-L6-v2, and is deployed as a live web app. The project is a stepping stone toward Siagnos, a future personal taste engine that will learn from reading behavior.

• •Opsis is a deployed content-based fanfiction recommender using embedding similarity and metadata overlap
• •Scrapes and processes 7,000+ AO3 fics with PostgreSQL, FastAPI, and sentence-transformers
• •Deliberately scoped down from a larger taste-engine vision to ship something usable in two weeks

Generated with AI, which can make mistakes.