SARYCH-LM Neural Observatory
SARYCH-LM Neural Observatory is a dark, cinematic research cockpit for my personal from-scratch 29.77M parameter language model project.
The goal was to make the project feel alive: not as a chatbot, but as a visual interface for exploring model facts, transformer architecture, training history, and a simulated Neural Playground terminal.
How I used Google Stitch:
- Connected Stitch through OpenCode using Stitch MCP
- Generated many visual concepts and screen variants
- Selected a dark neural observatory direction
- Ran factual correction passes to keep model details honest
- Exported the selected Stitch concept as HTML
- Rebuilt and polished the final site in Vite + React + TypeScript + Tailwind
- Deployed the final static site on GitHub Pages
The Neural Playground is UI simulation only — not live backend inference. SARYCH-LM is a personal research project: v0.3 is complete, while v0.4 instruction-tuning and synthetic-data infrastructure is still in progress.
Feedback on Stitch:
Stitch made rapid visual exploration genuinely fast and useful. The MCP workflow through OpenCode made prompt-driven iteration feel natural and close to my development process. Larger structural edits still benefited from manual frontend polish after export, but Stitch was excellent for finding the visual direction quickly.
Live site:
https://apartman36.github.io/sarych-lm-site/
X walkthrough post:
https://x.com/apartmanchik/status/2061514383713305020
Stitch project:
https://stitch.withgoogle.com/projects/7641294863595848179
Model repo:
https://github.com/Apartman36/sarych-lm
Website repo:
https://github.com/Apartman36/sarych-lm-site
(https://github.com/Apartman36/sarych-lm-site)Built with Google Stitch + OpenCode MCP.