# Research Microscope

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Franklin Alejandro Echeverria Baru](https://cerebralvalley.ai/u/FrankEBaru)
- **GitHub:** https://github.com/FrankEBaru/research-microscope
- **Demo video:** https://youtu.be/NpIZdZCemTY
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3957

Research Microscope is a reader for scientific papers that helps researchers understand them, not just find or summarize them.

Papers now arrive faster than we can read them: arXiv alone received over 40,000 new submissions in September 2026, and on a single day in October over 700 AI-written math manuscripts were released at once. AI helps us produce more papers; we believe it should also help us understand them. Terence Tao calls this "Math 2.0". We think every field deserves it.

How it works:
- Map of the field. Ask a question in plain words. We search ~8M papers (SPECTER2 embeddings + FAISS + the Semantic Scholar citation graph) and draw a map of the field by year, with groups named by Muse and papers ranked by closeness to your question, so you know where to start.
- Scan or Read. Scan gives the abstract, figures and conclusions; Read opens the full paper in a clean reader.
- Symbols. Hover any symbol and every use lights up, including inside equations; the margin quotes where the author defined it.
- Gap notes. Where the author says "it's easy to see", Muse writes a short explanation grounded in this paper and in passages from other papers (nearly 300,000 research passages), each line labeled AI-generated, From this paper, or From other papers.
- Live equations. Our engine reads an equation, and Muse Wonder writes the code for an interactive instrument under a strict contract. A separate verifier checks it against the paper's claim and a sandbox runs it. Today 9 of our 11 widgets pass.
- Explain. After a section, you explain it in your own words; Muse checks every claim against the author's sentences and other papers and answers Matches / Contradicts / No support / You missed, with quotes. On our test set of 137 hand-written claims the verdict was right 99% of the time.

Today we focus on computer science and math. Next: a research hub with weekly spaced questions on what you read, and connections across fields. Attribution: metadata from Semantic Scholar; papers from arXiv, always linked to the source.

## More from Meta Global AI Developer Hackathon

- [ProtoMind](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3954)
- [Research Navigator](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3955)
- [Road to 5K](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3956)
- [Second Opinion](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3958)
- [Tonelab Muse](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3959)
- [CloveaXD](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3960)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3957. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt