# NanoForensics

- **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:** [Prakash Khatri](https://cerebralvalley.ai/u/pkhatri)
- **GitHub:** https://github.com/0xPkhatri/NanoForensics
- **Demo video:** https://www.youtube.com/watch?v=QmRvxXcSrz0
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3835

NanoForensics is an AI-assisted scientific workbench that investigates why a simulated material breaks.

Instead of stopping at a plausible explanation, it proposes competing causes, tests controlled repairs through genuine atomistic simulations, and independently checks what the measured evidence supports.

Meta Muse Spark generates structured hypotheses, contributes experiment-planning judgments, and revises explanations after results arrive. LAMMPS—or a verified NumPy fallback—calculates atomic motion and mechanical response. A separate deterministic verifier evaluates numerical validity and scientific evidence; the model’s confidence cannot override its verdicts.

The dashboard lets users inspect atomic structures, compare stress–strain curves, run a complete investigation or explore individual tests, and follow a live execution console. Each action distinguishes live model responses from rule-based fallback and fresh simulations from saved results. Expandable records and exports preserve execution details and provenance.

In the published DEMO_501 example, restoring missing interior atoms increased measured fracture strain by approximately 47.85%, while the competing edge repair produced approximately −0.61% change. These findings apply to that small 2D simulation. The AI demonstration is a post-study rehearsal, rather than a new blinded evaluation.

Built with Next.js, FastAPI, SQLite, Muse Spark, and genuine molecular dynamics, NanoForensics explores how AI can make scientific investigation easier to understand, test, and audit.

AI proposes. Physics measures. Independent rules verify.

## More from Meta Global AI Developer Hackathon

- [Barnaby](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3832)
- [META VERSE](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3833)
- [HEARSAY-IV: Speech Transcription You Can Argue With | Meta Hackathon 2026](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3834)
- [Formwright](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3836)
- [barter.computer](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3837)
- [Trinetra](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3838)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3835. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt