Tetstack

Bro I am not good at this social media stuff 🥀

Good morning. @Tetstack

Good night, may our dreams come true. @Tetstack

The ragebait was good

I am on a streak, I suppose.

Good morning @Tetstack

Don’t become “unapologetically being yourself” 😂

Good night everyone. Connection is a bit slow.

bruh

So gemini is congested, groq is shit for this, mistral not working and so I will try to find another free API.

Haven't built the scrapper yet, maybe tomorrow or I will pull an all nighter

Update to the structure, the previous one was a prototype made back then.

here is the file structyre

┌─────────────────────────────────────────────────────────────┐
│                    YOUR PC (LOCAL)                          │
│  scraper.js  →  Playwright headless                         │
│    • visits CodePen / ShaderToy / Awwwards                  │
│    • extracts GLSL strings + render loops                   │
│    • writes DIRECTLY to Cloudflare D1 via wrangler CLI      │
│      (no local storage of scraped data)                     │
└────────────────────────────┬────────────────────────────────┘
                             │  wrangler d1 execute
                             ▼
┌─────────────────────────────────────────────────────────────┐
│                CLOUDFLARE D1 (CLOUD)                        │
│  Tables:                                                    │
│    scraped_shaders(id, source_url, glsl_code, js_snippet)  │
│    scrape_log(id, scraped_at, source, success)             │
│    site_state(id, current_html, version_hash, updated_at)  │
└────────────────────────────┬────────────────────────────────┘
                             │
                             ▼
┌─────────────────────────────────────────────────────────────┐
│         CLOUDFLARE WORKER (cron: daily 6AM UTC)             │
│                                                             │
│  1. Read latest scraped_shaders from D1                     │
│  2. Read current_html from site_state                       │
│  3. Build prompt:                                           │
│       system: "You are a WebGL code generator..."           │
│       user:   current_html + scraped_shaders + brief        │
│  4. Call Gemini API (free tier)                             │
│  5. Validate: esbuild syntax check + size budget            │
│  6. Write new_html back to site_state                       │
│  7. Increment version_hash                                  │
└────────────────────────────┬────────────────────────────────┘
                             │
                             ▼
┌─────────────────────────────────────────────────────────────┐
│              STATIC HOST (Cloudflare Pages)                 │
│  index.html  ←  fetches from Worker endpoint                │
│  Polls /api/version every 60s                               │
│  If hash changed → reload / hot-swap iframe                 │
└─────────────────────────────────────────────────────────────┘

Here is the diagram

which means I will use the API and skip the making the LLM from scratch.

My pc won't handle running the LLM on my pc so I am going hybrid mode

SO anyways, some of you guessed it, but with little changes I will be making Artificial Neural Networks (ANNs), Large Language Models (LLMs) for code, and real-time graphics engines. A multi-part artificial neurological system that continuously learns, generates, and evaluates. The key challenge is balancing "meaningful content UI/UX designs while integrating rare Three.js techniques. Basically it means Generative WebGL Synapse, scrapes websites takes their designs learn on it and then make one based on the hundreds of websites it makes a page based on that progressively adjust with more websites launching. I would have used fruit fly brain but it have 116K nodes which is not enough for this, and also some other cases. @Tetstack