The Weaverarxiviq.com
arxiviq.comApplied ResearchLive arXiv RAG--:--
The Weaver · arxiviq.com

Search arXiv like it's your own corpus

A research retrieval assistant, live below. The same production stack we run for customers, pointed at arXiv. RAG Embeddings in production · browse the research registry →

Live arXiv search
Checking API…
  1. Corpus indexed
    arXiv abstracts → chunks in pgvector
  2. Embed query (two tiers)
    Full 384d vs Matryoshka t=8 + int8
  3. Retrieve top-k
    Side-by-side rank comparison
  4. Exam: org vs BGE/e5 panel
    arXiv MCQ scorecard, top-3 retrieval
Checking API…Full vs edge compare

Same query, two serving tiers side-by-side. Full uses the complete embedding (best retrieval). Edge applies Matryoshka prefix t=8 + int8 quant — the home-device story from EVIDENCE.md Family B.

arXiv MCQ exam

Each arXiv-domain question is answerable from one source chunk. A model scores a hit when its retrieval places the source chunk in the top-3. Live mode: scored against real org-embed vs BGE/e5 vectors over the arXiv corpus.

Corpus: harvested from arXiv (CS.CL / retrieval / embedding queries). Re-kickoff with pnpm harvest:arxiv then pnpm kickoff:orgs after the stack is up. Feedback routes misses to Data / Research via Orchestration.
Use this on your own papers
Design-partner seats

Design-partner seats get a dedicated workspace: your corpus, monthly retraining behind deploy gates, and direct roadmap input. First cohort, quarterly terms.