Developer Builds Live Fanfiction Recommender Using AO3 Metadata and Embeddings

A developer launched Opsis, a content-based fanfiction recommender system, after scoping down a broader personal taste engine project called Siagnos during a two-week internship window. Opsis uses sentence-transformer embeddings and AO3 metadata — including fandoms, relationships, and summaries — to find fics similar to a given work, without tracking user reading behavior. The system is built on FastAPI, PostgreSQL hosted on Neon, and deployed on Render, with a multi-page UI allowing users to submit a fic by ID, URL, or title and author. It currently indexes over 7,000 fanfiction works, predominantly from the My Hero Academia fandom, and can automatically scrape, embed, and store new fics on demand. The developer acknowledged the absence of formal accuracy benchmarks, citing a lack of labeled relevance data, while noting that informal testing showed top recommendations consistently scoring in a 50–70 percent blended similarity range.