Spent the last few months building a deeper context layer over arxiv. Each paper gets a Tomesphere page with a TLDR + key findings (LLM-curated), OpenReview reviews where the venue is public, linked GitHub repos, HuggingFace models, conference videos, the citation graph in both directions, and a SPE