World Turtle Farm
banner
worldturtlefarm.bsky.social
World Turtle Farm
@worldturtlefarm.bsky.social
5 followers 1 following 660 posts
NFT hedge fund with stable coin and utility coin. The project is guided by experiential information, and will continue to be creatively dynamic until the launch of the NFTs
Posts Media Videos Starter Packs
Pinned
I’m really enjoying both build in public and build in private 😛 making all the AI tools work together is a crazy productivity boost! Also TIL what dogfooding means in software lol
MONITORING (Optional)
📊 Prometheus (512MB) - Port 9090
📈 Grafana (512MB) - Port 3000

MESH NETWORKING (Optional)
🌐 Mosquitto MQTT (128MB)
🔗 LoRa Mesh Bridge (256MB)
DOCUMENT PROCESSING
📄 Apache Tika (512MB) - PDF/document parsing - Port 9998
🔤 Tesseract (512MB) - OCR - Port 8000
AI MODELS (llama.cpp)
🧠 Phi-2 Q4 (1.6GB) - Text analysis - Port 8080
👁️  MobileVLM V2 (1.4GB) - Vision - Port 8080
🎤 Whisper Tiny (75MB) - Speech - Port 9000
DATABASES
🐘 PostgreSQL 15 (512MB) - Port 5432
🔍 Qdrant Vector DB (512MB) - Port 6333
⚡ Redis Cache (256MB) - Port 6379
STORAGE (GARAGE) ← NEW!
🚗 Garage S3-compatible
📊 Memory: 150MB (vs MinIO's 512MB - 362MB saved!)
💾 Buckets: atlas-data, atlas-documents, atlas-code, atlas-temp
📍 S3 API: 3902, Admin: 3901
🌐 localhost:3902 (S3), localhost:3901 (Admin)
BACKEND (RUST) ← NEW!
🦀 Rust + Axum
📊 Memory: 128MB (vs Python's 1GB - 872MB saved!)
⚡ Performance: 10x faster, 50k+ req/s
📍 Port: 8000
🌐 localhost:8000
#teamTinyStack

🏗️ CURRENT ATLAS STACK

━━━━━━━━━━━━━━━━━━━━━━

FRONTEND
⚛️  React 18 + Webpack 5 + TailwindCSS + shadcn/ui + Lucide icons
📍 Port: 5001
🌐 localhost:5001
🖥️ EDGE SERVERS (MicroK8s)
├─ Cache models
├─ Process documents
├─ Relay to mesh
└─ Regional hubs

📱 PINEPHONE (Multipass VMs)
├─ 🐢 Turtle NFT identity
├─ Local AI processing
├─ Mesh networking
├─ Offline-first
└─ Community-powered
Original 2020 architecture used OpenYurt and microK8s with multiPass #leelooMultipass #geminiCroquets

☁️ CLOUD (OpenYurt Control Plane)
├─ Manage all devices
├─ Deploy workloads
├─ Turtle NFT registry
└─ Model training
📱 PINEPHONE (Offline):

Ultra-Lightweight Models:
🤖 Phi-2 Q4 (1.6GB) ⭐ PRIMARY
🤖 TinyLlama Q4 (600MB) - Backup
👁️ MobileVLM Q4 (1.2GB) - Vision
🗣️ Whisper Tiny (75MB) - Speech

Total: ~3.5GB on device
RAM: ~2GB loaded
Battery: 6-12 hours
Accuracy: 92% (offline!)
🖥️ EDGE (Inference):

Quantized Models:
🤖 Mistral 7B Q4 (2.8GB) - General
👁️ LLaVA 13B Q4 (7GB) - Vision/OCR
💻 CodeLlama 13B Q4 (7GB) - Code
🏥 Meditron 7B Q4 (3.8GB) - Health
🗣️ Whisper Medium (1.5GB) - Speech

Total: ~30GB cached
Serves: 100-1000 devices each
☁️ CLOUD (Training):

Full-Size Models:
🤖 Llama 3.1 70B (140GB) - Training base
🤖 Mixtral 8x22B (176GB) - Mixture of experts
🤖 GPT-4 (API) - Validation
🎨 Stable Diffusion XL (6.9GB) - Turtle NFTs

Purpose: Train → Fine-tune → Quantize → Distribute
I’m a predictably unpredictable self organizing principle that operates on island time - who am I? #calledIt #strongAndDeep #strangeAttractor
10: ticket assignment swapping to split the work clock on tickets
11: committing someone else’s code to make git think it’s your code
8: building something wrong on purpose just to get something done
9: being criticized for making my own tickets
4: keeping quiet and getting my work done but then being criticized for not speaking up
5: being asked to do work that doesn’t make sense or is scoped incorrectly
6: being asked to re-do work that was already done
7: being asked to do work for someone who knows how to do what is being requested
These features are learnings from my career history:
1: being assigned too many or too little tickets on purpose
2: changing ticket scope with too little or too much information to impact velocity
3: suggesting improvements and being criticized for trying to improve processes
Something about Taoism Goats and Oat Milk
Worst experience of modern software development: PM and managerial ticket velocity games #theseMetricsWereDoctored
Engineers, never worry about being assigned tickets ever again 😜 Automated velocity and proof of work