# Versus

- **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:** [VER SUS](https://cerebralvalley.ai/u/vseurs)
- **GitHub:** https://github.com/Versus-DEV/Mequa
- **Demo video:** https://youtu.be/Q0XYirJebws
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2854

I built Mequa after watching tissue-mechanics researchers lose days to glue code. Getting from a microscope image to a single number meant stitching together one-off scripts — segmentation in one tool, mechanics in another, systems models in a third — and the only thing holding it all together was a folder of scripts that one person on the team understood. And when people reached for AI assistants, they got confident-sounding answers with made-up numbers, which in science is worse than no answer at all.

Mequa puts the whole workflow on one screen, and it runs on an ordinary laptop with no GPU. You feed it microscopy images, 3D stacks, point-cloud CSVs, or Simulink/Scilab models, then ask questions about your experiment in plain English.

The important part: the language model never does the math. A local Llama model figures out what you're asking and calls real numerical solvers as tools. Those solvers exist in three independent implementations — Julia, Rust, and NumPy — and they're cross-checked against each other, so a bug in one can't slip through quietly. Load sweeps, strain relaxation, density zones, SPHI, failure models: every number you see comes from a verified computation, not from the model's imagination.

And because biomedical tools inevitably attract people asking for personal medical advice, I made the guardrails part of the architecture instead of a disclaimer. Three layers (rules, Llama Guard 3, and an intent classifier) check every request and every reply. Mequa will never give anyone a diagnosis or tell them what to do about their health — it's an instrument for datasets, cohorts, and simulations, and it stays that way.

## More from Meta Global AI Developer Hackathon

- [Autonomous](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2851)
- [Mr-Priyanshu-Tech-Rebel](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2852)
- [Safe House Vision & Response](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2853)
- [CareerForge Innovators](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2855)
- [MuseMeet](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2856)
- [Ayna](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2857)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2854. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt