Since August I’ve been running a scoring system that tracks and ranks every new AI-related paper published on arXiv. The goal is to detect genuine research events, things that move the field forward and filter out the usual noise and unsubstantiated uploads.

Each paper gets evaluated through a scoring algorithm. The system then stores embeddings with structured fields like problem, approach, solution, and result for later retrieval.

Search runs on a hybrid model: it combines semantic embeddings with metadata like title, topic, author, and extracted keywords. Queries return the most relevant and technically meaningful papers for a given concept.

So far, the data (since August) looks stable and the scoring aligns surprisingly well with expert-curated lists. All of this is available through a free web interface and API.

Is this useful for everyday