P-008 · Case study
OPEN SOURCERumor Discovery - Event Recommendation Engine
Event recommendation engine built on weighted hybrid scoring: 5D vector embeddings, cosine similarity, and role aliasing so a CEO and a Founder read as the same person.
PLATE 01 · Rumor Discovery — interface
Problem
Event listings are one-size-fits-all; nothing scored how well an event actually fit a person
Solution
A weighted scoring formula over 5D vector embeddings and cosine similarity
Impact
Every user-event pair gets a score and a readable explanation
Users
25 user personas across 6 industries with 40 curated events
More detail
An event recommendation engine that matches users to events using 5D vector embeddings, cosine similarity, and weighted hybrid scoring. The scoring formula combines vector similarity, audience fit, history, and location weighting. Role aliasing maps equivalent titles (CEO↔Founder, VC↔Investor), explanations are pre-computed for 1,000 user-event combinations, and the LLM layer is multi-provider. Built with Next.js and TypeScript; tests cover cosine similarity validation, score distribution analysis, and pipeline integration.