Landing page
Time Series Forecasting
Artificial Neural Network
Clustering
Random Forest
Results Summary

Inspiration

As a scientist, I often see colleagues struggling with data analysis tasks such as neural networks, time-series forecasting, or clustering. Many of them are not familiar with statistics or programming, making it difficult to use tools like R or Python. While MS Excel works for simple statistics, advanced packages like SAS or SPSS require expensive licenses and steep learning curves.

The idea behind NetCraft AI was clear: 👉 Deliver machine learning capabilities through a simple, intuitive, client-side web application so that anyone can perform robust neural network prediction, clustering, or time-series forecasting in just a few clicks.

What it does

🧠 Neural Network Predictor

Custom multilayer perceptron in TypeScript
Regression & classification support
Configurable architecture (layers, neurons, activations)
Real-time training progress (loss curves, epoch tracking)
Evaluation metrics: MSE, MAE, R², accuracy, confusion matrix
Built-in normalization and preprocessing
Model serialization for persistence & deployment

🌳 Random Forest Classifier

Decision tree ensembles with bagging
Feature importance via impurity decrease
Out-of-bag validation (OOB scoring)
Regression & classification support
Configurable parameters (trees, depth, sampling ratios, seed)
Batch predictions with confidence scores
Flexible feature sampling strategies

📈 Time Series Forecasting

Moving Average with adjustable window sizes
Exponential Smoothing (simple & Holt’s method)
Linear trend & polynomial regression forecasting
Metrics: MAE, RMSE, MAPE, R²
Confidence intervals & uncertainty quantification
Automatic frequency detection & preprocessing
Train/test split for validation

🎯 Clustering Analysis

K-means clustering (with elbow method & k-means++ init)
Self-Organizing Maps (SOM) with multiple topologies
Interactive U-Matrix visualization
Metrics: silhouette score, inertia, topographic error
PCA for dimensionality reduction
Built-in normalization & standardization

📊 Data Management

CSV upload with drag-and-drop & validation
Auto column type detection (numeric, categorical, datetime)
Preprocessing for time-series gaps & validation
Data preview with stats & column selection
Demo datasets (Iris, Sales, Temperature series)
Data export to CSV

🎨 Modern UI/UX

Responsive Tailwind CSS design, mobile-ready
Smooth Framer Motion animations
Accessibility-first with ARIA support
Dark/light/system themes
Professional dashboard layout with sidebar nav
Interactive Chart.js visualizations
Loading states & progress indicators

🌍 Internationalization (i18n)

Bilingual UI (English & Ukrainian, full coverage)
Lazy loading of translations for performance
Preserves technical terms (e.g., RMSE, Random Forest)
Locale-aware formatting for dates/numbers
Accessibility announcements for language switch

💾 Export & Persistence

Save/load models as JSON
Export predictions, clusters, forecasts, feature importance to CSV
LocalStorage project state with migration support
Reports in JSON with metadata
Export charts as images

🔄 Data Migration & Compatibility

Backward compatibility & graceful fallbacks
Seamless data preservation during updates
Status tracking & error handling
Safe cleanup of old keys

How we built it

Vibes → skeleton app, QA tests, targeted improvements
Steering → steering docs & roadmap
Specs → rebranding, UI design, added ML models, bilingual support
Hooks → auto docs update & cleanup after testing

Tech stack → React 18, TypeScript 5.2+, Tailwind CSS 3.3+

Challenges we ran into

Daily/monthly request limit errors
Kiro freezing on commands like npm run dev
Occasional infinite loops during error fixing
Misinterpretations of prompts causing incorrect builds
Pre-revert, difficult to roll back bad changes

Accomplishments we’re proud of

Built a fully functioning client-side ML platform with almost no manual coding
Implemented advanced ML workflows in a short timeframe
Delivered bilingual support with smooth UI/UX