🌐 Portfolio • 💼 LinkedIn • 📧 Email • 🎓 Resume
| Metric | Value |
|---|---|
| Projects Shipped | 4 production systems |
| ML Model R² | 99%+ (LoadIQ ensemble) |
| MAPE Accuracy | < 1.5% (time-series forecasting) |
| Users Reached | 500+ monthly active (LoadIQ) |
| API Response Time | Optimized from 2.3s → 380ms |
| Data Processed | 6 years of half-hourly grid data |
| GitHub Stars | 50+ across repositories |
| Certifications | 15+ (Deep Learning, AI, Backend, Cloud) |
LoadIQ predicts real-time electricity demand across 4 countries (UK, USA, Germany, India) using advanced ML ensemble methods trained on 6 years of half-hourly transmission-level grid data (180K+ data points, zero null values).
| Achievement | Details |
|---|---|
| ML Model Performance | R² > 99% • MAPE < 1.5% • Chronological train/test split |
| Ensemble Architecture | LightGBM (60%) + XGBoost (40%) • Ridge regression fallback |
| Feature Engineering | 43-55 features • Lag windows (1h-168h) • Rolling statistics |
| API Optimization | Response time: 2.3s → 380ms (83% reduction) • Batch prediction support |
| Frontend UX | Lazy loading • Skeleton screens • Three.js WebGL globe |
| Deployment | Docker containerization • Nginx reverse proxy • Firebase Hosting |
| Reliability | Connection-refused guard • WebGL context recovery • Error fallback system |
┌─────────────────────────────────────────────────────────────┐
│ LoadIQ Platform │
├─────────────────────────────────────────────────────────────┤
│ │
│ Frontend (React 18 + TypeScript) │
│ ├─ Lazy-loaded pages (code splitting) │
│ ├─ Skeleton-first UI (instant perceived load) │
│ ├─ Three.js WebGL globe (4-country visualization) │
│ └─ Recharts forecasting dashboard │
│ │
│ ↓ Unified API Layer (Connection Guard) │
│ │
│ Backend (FastAPI + Uvicorn) │
│ ├─ Model1.py (Port 8001: DE+LU Load Prediction) │
│ ├─ Model2.py (Port 8002: GB Load Prediction) │
│ ├─ Health checks & request validation │
│ └─ Batch prediction (24-168 steps ahead) │
│ │
│ ML Inference Layer │
│ ├─ LightGBM Model (primary, 60% weight) │
│ ├─ XGBoost Model (secondary, 40% weight) │
│ ├─ Ridge Regression (fallback) │
│ └─ Feature preprocessing (median fill for missing) │
│ │
│ Data Layer │
│ ├─ Training: 50.4K rows (DE+LU, 60-min intervals) │
│ ├─ Training: 100.8K rows (GB, 30-min intervals) │
│ ├─ Time range: 2015-2020 (6 years of grid data) │
│ └─ Features: 45-46 columns per dataset │
│ │
│ Monitoring & Logging │
│ └─ Firebase Firestore (exception logging, analytics) │
│ │
└─────────────────────────────────────────────────────────────┘
| Decision | Impact |
|---|---|
| Demand-based page loading | Reduced bundle size by 45% |
| Skeleton-first UI | Users perceive <500ms load time |
| Median-fill strategy | Preserves model integrity vs. zero-fill |
| TimeSeriesSplit CV | Prevents data leakage in temporal data |
| Ensemble blending | Smooths variance at peak transitions |
| Dual-server architecture | Country-specific models optimize for regional patterns |
| Connection fallback | Graceful degradation when backend offline |
Frontend: React 18 • TypeScript • Three.js • Recharts • Firebase Hosting
Backend: FastAPI • Uvicorn • Python 3.10+ • Nginx
ML Models: LightGBM • XGBoost • Ridge Regression • Scikit-learn
Data: Pandas • NumPy • ENTSO-E Transparency Platform
Infrastructure: Docker • Firebase Auth & Firestore • Google Cloud Platform
Model Performance (Chronological Test Set):
├─ LightGBM: R² = 0.992 • MAPE = 1.1%
├─ XGBoost: R² = 0.985 • MAPE = 1.8%
├─ Ridge: R² = 0.935 • MAPE = 3.8%
└─ Ensemble: R² = 0.993 • MAPE = 1.4% ✓
API Performance:
├─ Single prediction: 380ms (avg)
├─ Batch (24-step): 2.1s
└─ p95 latency: 650ms
# Clone & setup
git clone https://github.com/Parth-S-Mhatre/LoadIQ.git
cd LoadIQ
# Backend (Terminal 1)
cd Backend && pip install -r requirements.txt
python Model1.py # Port 8001
python Model2.py # Port 8002
# Frontend (Terminal 2)
cd energy-analytics && npm install && npm start
# Opens http://localhost:3000See LoadIQ README for full setup & API documentation.
Tech: Streamlit • Scikit-learn • Plotly • Python
- Predictive ML system forecasting student academic outcomes on 1000+ records
- Accuracy: 85% with comprehensive cross-validation
- Feature engineering: standardization, one-hot encoding, polynomial features
- Real-time prediction API with personalized recommendations
- Repository
Tech: Python • PyTorch • TensorFlow • Scikit-learn
Comprehensive collection demonstrating breadth across ML domains:
| Project | Type | Highlight |
|---|---|---|
| Reinforcement Learning | RL | Self-driving car simulation, Lunar Lander, Q-Learning with Pygame |
| Customer Churn Prediction | Classification | Ensemble: Logistic Regression, Random Forest, SVM, Gradient Boosting |
| Car Price Predictor | Regression | Linear Regression + Flask deployment |
| Titanic Survival | Classification | 81% accuracy • Complete ML pipeline |
| Mumbai House Prices | Regression | Advanced feature engineering |
Each project includes: EDA, preprocessing, model selection, cross-validation, evaluation metrics.
Tech: Java • Spring Boot • Maven • REST APIs
- API Basics Mastery: REST principles, HTTP methods, JSON, async operations
- Journal Application: Building production-grade backend with Spring Boot
- Hands-on Java exercises demonstrating DSA and core CS concepts
- Java DSA
Tech: Python • Algorithms • Data Structures
Clean examples demonstrating Python best practices and algorithmic thinking.
Contributions:
- ✅ Resolved API rate-limit instability via 30-second backoff strategy
- ✅ Reduced failed API calls by 60% during peak load
- ✅ Built modular React landing-page components
- ✅ Improved component library for responsive design
- ✅ End-to-end integration testing & deployment
Core Competencies: Time-Series Forecasting • Ensemble Methods • Feature Engineering • Model Optimization • Reinforcement Learning
Core Competencies: RESTful API Design • Microservices • Error Handling • Request Optimization
Core Competencies: Component Architecture • Lazy Loading • Skeleton Screens • WebGL Optimization
Core Competencies: Containerization • Cloud Deployment • Monitoring & Logging
Bachelor of Technology in Computer Engineering (AI & Data Science Specialization)
Pillai College of Engineering, Mumbai | Aug 2023 - June 2027
- ✅ IBM Deep Learning with PyTorch, Keras and TensorFlow (March 2026)
- ✅ Machine Learning Specialization — DeepLearning.AI, Stanford (June 2025)
- ✅ Deep Learning with PyTorch (IBM, Feb 2026)
- ✅ Deep Learning with Keras & TensorFlow (IBM, Dec 2025)
- ✅ Introduction to Neural Networks & PyTorch (IBM, Jan 2026)
- ✅ AI Capstone Project with Deep Learning (IBM, Mar 2026)
- ✅ AI Agents and Agentic AI with Python (Vanderbilt, Aug 2025)
- ✅ API Basics: REST, HTTP, JSON (Scrimba, April-June 2026)
- ✅ Google Cloud Fundamentals & Hands-on Labs
- Plus 6 additional courses in Cloud, ML, and Backend Development
Total: 15+ Professional Certifications
Writing technical deep-dives on key projects:
- Building AI-Powered Energy Forecasting at Scale — LoadIQ architecture, challenges, and lessons learned
- Production ML: From 2.3s to 380ms API Response Time — Optimization techniques and profiling
- Feature Engineering for Time-Series: The LoadIQ Approach — 43-55 features, lag windows, temporal encoding
- GitHub: @Parth-S-Mhatre
- LinkedIn: @parthmhatre41
- Email: parth.mhatre4141@gmail.com
- Portfolio: parth-s-mhatre.netlify.app
Last Updated: June 2026
Building the future of AI-powered systems, one project at a time.