# MolAstria

- **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:** [Jiasheng Fu](https://cerebralvalley.ai/u/Felix723)
- **GitHub:** https://github.com/FelixJiashengFu/chem-research-molastria
- **Demo video:** https://www.youtube.com/watch?v=BzQbpSZR0gw
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4297

MolAstria (https://molastria.com) is a computational chemistry atlas I built as a chemistry student, for a problem I hit every week: the field's databases, software, methods and papers are scattered, and most of them exist only in English.

- Closed-loop AI Lab for electrolyte solvent discovery: Muse Wonder 1.0 proposes candidate molecules as SMILES with a design rationale; RDKit.js filters them in the browser (allowed elements, size, hazardous groups, InChIKey de-duplication, PubChem novelty check); the lab writes Rowan scripts that screen candidates with Meta's open-weights UMA model and then wB97M-V DFT, imports the results, scores their reliability (UMA-S/M disagreement, applicability domain) and spends the DFT budget on the ionization-potential / electron-affinity Pareto front. Each round's prompt carries the front, the rejected molecules and the predictions DFT overturned back to the model. Demo values are labelled as synthetic until real results are imported.
- AI research assistant on every page: Muse Spark 1.3 and Muse Wonder 1.0 through the Meta Model API (or DeepSeek / any OpenAI-compatible endpoint). "Ask the catalogue" retrieves site entries first and answers with citations.
- Search over 1,500+ deduplicated resources with Chinese-English term matching, a resource network graph, a 19-chapter Chinese guide, and PreprintRadar across 12 preprint servers.
- Chinese / English UI, light and dark themes, a Windows / macOS desktop app, and Pro accounts with weekly AI-picked papers.
- Meta × Chemistry page (https://molastria.com/meta/): Muse Spark 1.3 reads structure images into SMILES that RDKit checks one by one; a generator of ready-to-run fairchem + ASE scripts for Meta's UMA (adiabatic IP/EA with UMA-S/UMA-M cross-checks, conformer search, MD); and a catalogue of FAIR Chemistry's open models and datasets. Desktop app 0.1.5 runs these UMA jobs with the user's own Python and GPU and imports IP/EA straight into the AI Lab.

## More from Meta Global AI Developer Hackathon

- [LawnLens](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4294)
- [toba_apex](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4295)
- [Cruyff](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4296)
- [SAMREEN.J](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4298)
- [Delite99](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4299)
- [Drik.Ai](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4300)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4297. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt