# No-Code Vision Suite: Dynamic CNN Classifier Studio

A production-grade, full-stack, no-code deep learning dashboard that empowers users to build, configure, train, and test custom Convolutional Neural Networks (CNNs) directly from an interactive user interface. The platform bridges the gap between raw dataset operations and network training matrices by providing real-time telemetry streaming and automated evaluation pipelines.

---

🚀 Key Architectural Features

• Dynamic Network Architecture Sandbox: Adjust structural model layouts on the fly—including custom layer depth (1 to 5 Conv Blocks), base feature channel widths ($16, 32, or 64 \text{ ch}$), Batch Normalization layers, and variable Dropout regularizations.
• Synchronized Telemetry Logging: Features a persistent Experiment Tracking Log connected directly to an underlying SQLite database backend. State mutation hooks ensure historical tracking rows instantly sync when datasets are modified or cleared.
• Server-Sent Events (SSE) Training Stream: Real-time epoch-by-epoch loss tracking is broadcast from the Python back-end directly to the UI canvas without manual page-polling overhead.
• Mathematically Hardened Evaluation Engine: Processes validation holdouts by computing a true Multi-Class Macro F1-Score derived explicitly from an on-screen runtime Confusion Matrix matrix.
• Deterministic Configuration Guard: Includes absolute system seed locking across Python, NumPy, and PyTorch frameworks to maintain reproducible experimental baselines when evaluating identical training matrices.

---

🛠️ System Technology Stack

Ecosystem Layer	Technologies Utilized
Frontend Layout Core	React 18+, TypeScript, Vite, Tailwind CSS, Lucide Icons
Backend Core Engine	FastAPI, Python 3.10+, PyTorch, Torchvision, Uvicorn
Database & ORM	SQLite, SQLAlchemy Core
Data Normalization	Pillow (PIL), NumPy

---

📂 Repository Directory Structure

no-code-ml-app/
├── frontend/                # Vite + React Client SPA
│   ├── src/
│   │   ├── components/      # ClassManager, ImageUploader, Trainer, Predictor, ExperimentTracker
│   │   ├── App.tsx          # Main Root Workspace Orchestrator
│   │   └── main.tsx         # Virtual DOM Application Bootstrapper
│   ├── index.html           # Anti-Flash Fixed Canvas Entry Node
│   └── package.json
└── backend/                 # FastAPI Application Core
    ├── app/
    │   ├── main.py          # Routing Layouts and API Definitions
    │   ├── models.py        # SQLAlchemy Schema & Dynamic PyTorch CNN Subclasses
    │   ├── ml_engine.py     # Hardened Macro F1 Training Loops & Inference Pipelines
    │   └── database.py      # Session Handlers and Engine Connections
    └── requirements.txt     # Python Dependencies Matchlist

⚙️ Local Installation & Startup Guide

Follow these instructions to run the application in your local environment. Open two separate terminal windows to host your execution loops:

1. Backend Infrastructure Configuration (Terminal 1) Navigate to the backend repository root, establish a clean virtual environment space, and activate the local server context:

   Navigate to the backend directory

   cd backend

      python3 -m venv .venv
   

   Activate the environment path wrapper

   On macOS or Linux:

   source .venv/bin/activate

   On Windows (Command Prompt):

   .venv\Scripts\activate

   Install essential deep learning and API framework binaries

   pip install -r requirements.txt

   Launch the live-reloading Uvicorn development loop

   uvicorn app.main:app --reload --host 127.0.0.1 --port 8000

2. Frontend Layout Compilation (Terminal 2) Navigate to your frontend repository root block, download node engine modules, and boot up the asset building pipelines:

   Navigate to the frontend directory

   cd frontend

   Download and map node dependencies

   npm install

   Run the local Vite development instance

   npm run dev

Once successfully loaded, navigate to http://localhost:5173 inside your web browser to interact with your dashboard space!

Thank You!