Aviral GargArxivian · Browser edition
Research notes.
Answers with evidence.
Add a few notes, ask a question, and follow each answer back to its source.
Runs in this tab. No account, API key or upload to a server. Notes disappear on reload.
Your research notes
Select a note title to read or edit it. Plain text or Markdown. Up to 1 MB per import and 200,000 characters in the collection. Sample notes are written for this demo.
Add a note by hand
What would you like to understand?
This edition uses BM25 keyword retrieval and quotes matching sentences. It does not call an LLM or compute embeddings.
Answer from your notes
Extracted sentences, with citations. Relevance scores rank passages; they are not confidence probabilities.
From the original RAG prototype to this browser edition
The original Arxivian prototype combined a React/TypeScript interface, Python APIs on AWS, OpenAI embeddings, a FAISS vector index and LLM answers grounded in retrieved context. Its cloud services were retired in July 2026.
This portable edition keeps the collect → retrieve → inspect workflow available on my portfolio. It splits notes into overlapping passages, ranks them with BM25 and extracts matching sentences with source citations. It does not recreate the original semantic search or generative backend.
Compare different questions, remove a note and inspect how the evidence changes. Keyword matching can miss paraphrases, and a relevant passage can still be incomplete. Always read the cited text.