stack.txt

Treść wiadomości Gemini
================================================================================
TECHNOLOGY STACK DESCRIPTION FOR THE "INTI CHECK" PROJECT
PROJECT NAME: IntiCheck - AI Medical Assistant
DESCRIPTION: A discreet web application utilizing artificial intelligence to
classify skin lesions based on photos. The application provides
AI classification along with first-aid advice and medical
recommendations.

================================================================================

BACKEND (SERVER)
================================================================================

FRAMEWORK: Flask (Python Web Framework)

Minimalist, lightweight framework for building web applications

Route management (HTTP paths)

Serving static files (HTML, CSS, JS, manifest.json, sw.js)

Port: 7860 (Configured for Hugging Face Spaces)

FLASK EXTENSIONS:

Flask-CORS: Handling Cross-Origin Resource Sharing (frontend ↔ backend communication)
Allows the frontend to support API requests from various sources

MACHINE LEARNING & IMAGE PROCESSING:

TensorFlow: Deep learning library for loading and running the model

Keras: High-level neural networks API (part of TensorFlow)

Model: inti_check_v5_pro.h5 (trained neural network model)

Architecture: Convolutional Neural Network (CNN)

Input: Images resized to 224x224 pixels

Output: Classification into 6 skin lesion categories

Categories: Fungal infection, Nail problems, Pigmented lesions,
Viral infections, Scabies/Bites, Acne/Purulent lesions

Format: H5 (HDF5 - Hierarchical Data Format)

SUPPORTING LIBRARIES:

NumPy: Numerical processing and array manipulation
Converts images into numerical matrices

Pillow (PIL): Image processing and manipulation
Reading files, resizing, format conversion

Werkzeug: WSGI (Web Server Gateway Interface) utilities
Handling file uploads and HTTP error handling

API ENDPOINTS:
POST /predict - Accepts a photo, returns classification with:
* Category (text)
* Description (what is happening)
* First-aid advice (doraźnie)
* Doctor recommendation
* Confidence score in %

================================================================================
2. FRONTEND (USER INTERFACE)
HTML5:

Semantic document structure

Meta tags for responsiveness (viewport, charset UTF-8)

PWA (Progressive Web App) integration

Progressive Enhancement (works without JS)

CSS3:

CSS Variables (--primary-bg, --accent-green, etc.) - fully responsive

Glassmorphism: Frosted glass effect using backdrop-filter blur(16px)

CSS Grid and Flexbox: Flexible, mobile-first layout

Linear Gradients: Beautiful color transitions (teal → dark green)

CSS Animations: Smooth transitions (0.3s)

Responsive Design: Breakpoints for mobile/tablet/desktop

Components:

Hero section with tags and headers

FAQ Section (Accordion)

Image upload card (glass card)

Photography instructions

HIPAA/Agreement compliance checkbox

JAVASCRIPT:

Event listeners for dropzone and file input

Fetch API: Asynchronous communication with the backend

FormData: Image conversion to binary POST request

DOM manipulation: Dynamic UI updates

Image preview: Previewing the photo before sending

FAQ Toggle: Expanding/collapsing FAQ sections

Error handling: Handling network and validation errors

GOOGLE FONTS:

Outfit (weights: 300, 400, 600, 700)

Modern, clean sans-serif font

================================================================================
3. PROGRESSIVE WEB APP (PWA)
MANIFEST.JSON:

Web app declaration

Metadata: Name, description, theme colors

Icons: 192x192 and 512x512 pixels (PNG)

Display mode: "standalone" (looks like a native app)

Start URL: /index.html

Theme color: #a3e635 (green accent)

Background color: #0f3f38 (teal)

SERVICE WORKER (sw.js):

Background script enabling offline functionality

Cache API: Resource buffering

Offline browsing (if model is pre-loaded)

Push Notifications: Potential notification support

Background Sync: Background data synchronization

Add-to-Home-Screen installation support

BROWSER SUPPORT:

Conditions: Chrome, Edge, Firefox, Safari (modern browsers)

Cache Storage API

IndexedDB (optional for storing results)

================================================================================
4. DOCKER & DEPLOYMENT
DOCKERFILE:

Base Image: python:3.11-slim (lightweight Python 3.11 version)

Size: ~300MB (without the ML model)

Workdir: /app (working directory in the container)

Steps:

Copying requirements.txt

Dependency installation: pip install --no-cache-dir

Copying the entire project

Creating a folder for uploads: /app/uploads

Port exposure: 7860

CMD: Running server.py

HUGGING FACE SPACES:

Integration: Dockerfile is optimized for HF Spaces

Free hosting for public ML projects

Auto-scaling

Port 7860 (default for HF)

REQUIREMENTS.TXT:

flask: Web framework

flask-cors: CORS support

tensorflow: ML framework (≈900MB after installation)

pillow: Image processing

numpy: Numerical computing

werkzeug: WSGI utilities

================================================================================
5. KNOWLEDGE BASE AND BUSINESS LOGIC
KNOWLEDGE_BASE:

Python dictionary with 6 categories

Structure:
{
index: {
"category": string,
"description": string,
"first_aid": string,
"doctor": string (specialist recommendation)
}
}

Each category contains:

Medical description (what the classification means)

First-aid advice (immediate steps)

Recommended doctor specialty with emoji

Dynamically linked with the ML model results

DATA FLOW:

Frontend: User uploads photo (POST /predict)

Backend: Flask receives the file

Processing: Pillow → resize to 224x224 → normalization

AI: TensorFlow/Keras → predict() → vector of 6 outputs

Interpretation: argmax() → category index, max() → confidence %

Enrichment: Attaching text from KNOWLEDGE_BASE

Response: JSON with results + advice

Frontend: Displaying results to the user

================================================================================
6. SECURITY AND PRIVACY
HIPAA COMPLIANCE (Healthcare Privacy):

HIPAA-compliant checkbox on the frontend

Informational (not medical) warnings

Disclaimer: "This is not a medical consultation"

IMAGE UPLOADS:

In-memory processing: processed in RAM without saving to disk

Conversion to RAM buffer

Alpha channel removal (RGBA → RGB)

Normalization: /255.0 (scaling to 0-1 range)

CORS:

Flask-CORS enabled for secure frontend ↔ backend communication

================================================================================
7. GENERAL ARCHITECTURE
PROJECT STRUCTURE:
/
├── server.py             (Backend - Flask + TensorFlow)
├── index.html            (Frontend - UI)
├── sw.js                 (Service Worker)
├── manifest.json         (PWA config)
├── inti_check_v5_pro.h5  (ML Model - 900MB+)
├── requirements.txt      (Python dependencies)
├── Dockerfile            (Container config)
└── notes.txt             (documentation)

COMMUNICATION:
Browser ↔ (HTTP/HTTPS) ↔ Flask Server
↔ (Local API) ↔ TensorFlow Model

PERFORMANCE:

Inference: ~2-5 seconds (depending on CPU/GPU)

Image: Resize + normalization in RAM

Model cache: Loaded once at startup (warmup)

================================================================================
8. SUPPORTING TECHNOLOGIES
PYTHON VERSION: Python 3.11
OPERATING SYSTEM: Linux (Docker container)
DATABASE: None (stateless application)
CACHE: Browser cache (Service Worker)
CDN: Google Fonts (for typography)

MODERN STANDARDS:

ES6+ JavaScript

CSS Grid/Flexbox

REST API

JSON format

Responsive Web Design

Mobile-First approach

Web Components (potential)

================================================================================
SUMMARY
The IntiCheck project is a modern web application combining:
✓ Deep Learning (TensorFlow/Keras for image classification)
✓ Responsive frontend (HTML5 + CSS3 + Vanilla JS)
✓ Progressive Web App (offline capability)
✓ Lightweight backend (Flask + Python 3.11)
✓ Containerization (Docker for easy deployment)
✓ Free hosting (Hugging Face Spaces)

Optimized for:

User privacy (In-memory processing, no DB)

Speed (Model loaded once, inference in 2-5s)

Accessibility (PWA, offline mode)

Mobility (App installable on the home screen)

================================================================================
Documentation Created: 2026-03-06
Version: 1.0