# Manuel baslatmak icin:
python -m uvicorn live_stream_analyzer.backend.app:app --host 127.0.0.1 --port 8080

# JSON-RPC 2.0 STDIO Ajan Baglantisi
python -m ultratribe.mcp
# veya
ultratribe-mcp

# Kendi Kendini Test Etme (Diagnostics & In-Memory Benchmark)
python -m ultratribe.mcp.server --test

{
  "tools": [
    {
      "name": "system_diagnostics",
      "description": "Donanim kaynaklarini (CUDA, VRAM, PyTorch, FlashAttention-2, BF16 destegi) denetler.",
      "inputSchema": { "type": "object", "properties": {} }
    },
    {
      "name": "benchmark_inference",
      "description": "Sentetik girdi ile bellek-ici ileri yayilim (forward pass) gecikme (latency), verim (throughput) ve tepe VRAM olcumu yapar.",
      "inputSchema": {
        "type": "object",
        "properties": {
          "batch_size": { "type": "integer", "default": 4 },
          "seq_len": { "type": "integer", "default": 64 },
          "iterations": { "type": "integer", "default": 5 }
        }
      }
    },
    {
      "name": "list_supported_studies",
      "description": "Sistemde kayitli fMRI calisma veri setlerini listeler (Algonauts 2025, BOLD5000, Wen 2017, Lebel 2023).",
      "inputSchema": { "type": "object", "properties": {} }
    },
    {
      "name": "get_atlas_regions",
      "description": "Kortikal HCP veya Subkortikal Harvard-Oxford ROI beyin bolge indekslerini dondurur.",
      "inputSchema": {
        "type": "object",
        "properties": {
          "atlas_type": {
            "type": "string",
            "enum": ["cortical_hcp", "subcortical_harvard_oxford"],
            "default": "cortical_hcp"
          }
        }
      }
    }
  ]
}

# Standart gelistirici kurulumu
pip install -e .

# Tam kurumsal kurulum (API, gorsellestirme ve test araclari dahil)
pip install -e ".[serve,viz,dev]"

import torch
from ultratribe.core import FmriEncoderModel
from ultratribe.config import TribeConfig

# 1. Pydantic v2 tabanli tip-guvenli yapilandirma
config = TribeConfig().model

# 2. Model baslatma
model = FmriEncoderModel(config)

# 3. Girdi tensörleri hazirlama (Batch, Time, Modality_Dim)
dummy_batch = {
    "subject_id": torch.tensor([0]),
    "video": torch.randn(1, 64, 64),
    "audio": torch.randn(1, 64, 32),
}

# 4. Hizli cikarim
with torch.inference_mode():
    cortex_output = model(dummy_batch)

# Cikti boyutu: (Batch: 1, Vertices: 20484, Timesteps: 64)
print(f"Cikti Tensör Boyutu: {cortex_output.shape}")

ultratribe/
|-- live_stream_analyzer/    # Canli YouTube ve Video 3D Beyin Analiz Uygulamasi
|   |-- backend/             # FastAPI, WebSocket, Nöral Kodlama Motoru, Explainer
|   |-- frontend/            # OLED True Dark WebGL (Three.js) 3D Beyin Arayuzu
|   `-- baslat.bat           # Tek tikla Windows baslatici
|-- core/                    # Sinir Agi ve Egitim Cekirdegi (FmriEncoderModel, FlashAttention-2)
|-- mcp/                     # Model Context Protocol Katmani (Tools & Resources)
|-- api/                     # REST & SSE Cikarim Sunucusu (FastAPI, Rate Limiting)
|-- data/                    # MMap Streaming Veri Yukleyici & Donusumler
|-- config/                  # Pydantic v2 Tip-Guvenli Sema & Hiperparametreler
|-- shared/                  # Sparse ROI Hesaplama, Saf Fonksiyonlar
|-- viz/                     # 3D Beyin Gorsellestirme
|-- demo.py                  # Tek-satir TribeModel yukleyici
|-- Dockerfile               # Multi-stage container tanimi
`-- docker-compose.yml       # API + Redis + Prometheus + Grafana servisleri

# 1. MCP Protokol Dogrulamasi
python -m ultratribe.mcp.server --test

# 2. Birim Testleri
python -m unittest discover -s tests -p "test_*.py"