Jishnu P G - Full Stack Developer & ML Specialist Portfolio

Welcome to my professional portfolio! This repository showcases my expertise in full-stack development, machine learning, and data science through detailed project documentation and case studies.

🚀 About Me

I'm a final-year BCA student at Yuvakshetra Institute of Management Studies (YIMS), Palakkad, with a passion for building intelligent solutions that solve real-world problems. My technical journey spans:

• Full-Stack Web Development (React, Next.js, TypeScript)
• Machine Learning & AI (TensorFlow, Keras, Scikit-learn)
• Backend Development (FastAPI, Python)
• Data Science & Analytics (Pandas, NumPy)

📋 Featured Project

SmartWatt AI - Physics-Informed Hybrid AI System for Residential Energy Estimation

Problem/Context: Solved the "Linear Load Paradox" in residential energy estimation where pure AI models failed on linear loads (<31% accuracy) and traditional calculators missed aging appliance degradation.

Solution: Engineered a Physics-Informed Hybrid AI system for appliance-wise energy estimation, achieving remarkable accuracy improvements across all appliance types.

Key Achievements

Appliance Type	Model Accuracy	Improvement	Key Insight
Refrigerators	94.5%	+14.2% vs. Pure AI	Captures efficiency degradation in complex loads
Air Conditioners	87.6%	+12.9% vs. Pure AI	Predicts duty cycle variations accurately
Ceiling Fans	82.9%	+52.2% vs. Pure AI	Recovers linear load accuracy through physics constraints
LED Lights	86.0%	+65.9% vs. Pure AI	Eliminates AI hallucination with physics enforcement

Linear Load Precision: Achieved 98% accuracy on linear load predictions by enforcing physics constraints (Efficiency = 1.0), ensuring explainable and trustworthy predictions.

Technical Architecture

• Dual-Inference Engine: Dynamically switches between:

  • Neural Networks for complex non-linear loads (AC, Refrigerator)
  • Physics-Enforced Logic Gates for linear loads (Fans, Lights, Iron)

• Multi-Output Regression: Predicts both:

  • Efficiency Factor (age/star degradation)
  • Effective Duty Cycle (compressor on/off time)

• KSEB Tariff Engine: Compliant with 2024-25 Kerala electricity billing standards with telescopic slab calculations

Technology Stack

Frontend:

• Next.js 14 (App Router)
• React 19
• TypeScript
• Tailwind CSS

Backend:

• FastAPI (Python)
• TensorFlow/Keras (Neural Networks)
• Scikit-learn (Machine Learning)
• Pandas/NumPy (Data Processing)

Database:

• PostgreSQL
• Supabase

ML Techniques:

• Physics-Informed ML
• Multi-Output Regression
• Neural Network Optimization
• Physics-Enforced Logic

Key Innovations

1. Hybrid Engine Architecture - Combines AI and physics for optimal predictions
2. Multi-Output Prediction - Captures hidden variables like thermal leakage
3. Physics-Constrained Learning - Prevents AI hallucination on deterministic loads
4. User Intent Override - Respects manual inputs while maintaining physics integrity

Project Metrics

• Prediction Accuracy: 82.9% - 94.5% across appliance types
• Linear Load Precision: 98% with physics constraints
• Aging Appliance Detection: Identifies 18-22% degradation in appliances >7 years
• Real-time Processing: Sub-second inference on FastAPI backend
• User Coverage: Designed for Kerala household energy patterns

📁 Repository Structure

Portfolio/
├── README.md                          # This file
├── LICENSE                            # MIT License
├── resume.md                          # ATS-friendly resume
├── index.html                         # Static portfolio page
├── styles.css                         # Portfolio styling
├── script.js                          # Portfolio interactivity
│
├── public/
│   ├── index.html                     # React app entry point
│   └── assets/
│       └── resume.html                # Embedded resume
│
├── src/
│   ├── App.jsx                        # Main React component
│   ├── App.css                        # App styling
│   ├── index.js                       # React DOM entry
│   ├── index.css                      # Global styles
│   │
│   ├── components/
│   │   ├── Navbar.jsx                 # Navigation component
│   │   ├── Hero.jsx                   # Hero section
│   │   ├── About.jsx                  # About section
│   │   ├── Projects.jsx               # Projects showcase
│   │   ├── Skills.jsx                 # Technical skills
│   │   ├── Contact.jsx                # Contact form
│   │   ├── Footer.jsx                 # Footer
│   │   ├── GitHubActivity.jsx         # GitHub integration
│   │   ├── EngineeringFocus.jsx       # Engineering focus
│   │   ├── ResumeDownloadButton.jsx   # Resume download
│   │   ├── EasterEgg.jsx              # Easter egg
│   │   └── [Component].css            # Component styles
│   │
│   └── pages/
│       └── CaseStudy.jsx              # SmartWatt case study
│
├── build/                             # Production build files
├── package.json                       # Project dependencies
├── .gitignore                         # Git ignore rules
└── FUTURE_UPGRADES.md                 # Planned improvements

🛠️ Technologies & Skills

Programming Languages

• Python, JavaScript, TypeScript, SQL, C, Java, PHP

Frontend Development

• React, Next.js, HTML, CSS, Tailwind CSS, Plotly.js

Backend & ML

• FastAPI, TensorFlow, Keras, Scikit-learn, Pandas, NumPy

Databases & Tools

• PostgreSQL, Supabase, Git, jsPDF

Specializations

• Physics-Informed Machine Learning
• Hybrid AI Systems
• Energy Analytics
• Full-Stack Web Development

🚀 Getting Started

Prerequisites

• Node.js (v16 or higher)
• Python (v3.8 or higher)
• npm or yarn

Installation

# Clone the repository
git clone https://github.com/JishnuPG-tech/Portfolio.git
cd Portfolio

# Install dependencies
npm install

# Start the development server
npm start

The portfolio will open at http://localhost:3000

Building for Production

# Create production build
npm run build

# Serve the build
npm install -g serve
serve -s build

📊 Key Project Insights

The Linear Load Paradox Problem

Traditional energy estimation faced a fundamental contradiction:

• Pure AI Models: Excellent at learning complex appliance behaviors but hallucinate on simple deterministic loads
• Physics Calculators: Perfect for static loads but fail to capture aging-related efficiency degradation

SmartWatt's Solution

By implementing a hybrid approach that:

1. Uses AI only for complex, non-linear appliances
2. Enforces physics laws for simple, linear appliances
3. Predicts hidden variables (degradation factors, duty cycles)
4. Respects user intent while maintaining scientific accuracy

Result: 52-66% accuracy improvement over pure AI baselines!

🎓 Key Learnings

• Physics-Informed ML is superior for engineering domains where domain knowledge is available
• Hybrid architectures can leverage strengths of both AI and traditional methods
• Explainability matters - users trust predictions they can understand
• Real-world constraints (aging appliances, user behavior) are crucial for practical systems

📞 Contact & Connect

• Email: jishnupg2005@gmail.com
• Phone: +91 8590731979
• Location: Palakkad, Kerala, India
• GitHub: @JishnuPG-tech
• LinkedIn: Connect with me

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• YIMS Palakkad - For educational support and resources
• Calicut University - Curriculum guidance
• Open Source Community - For excellent libraries and frameworks
• Energy Analytics Community - For inspiration and insights

📈 Future Roadmap

See FUTURE_UPGRADES.md for planned enhancements including:

• Mobile app for iOS/Android
• Real-time IoT device integration
• Advanced visualizations
• Community energy analytics dashboard
• Comparative analysis across regions

📝 Notes

This portfolio represents my current capabilities and interests. I'm continuously learning and exploring new technologies. Feel free to reach out for collaborations, feedback, or opportunities!

---

Last Updated: December 2025
Portfolio Version: 2.0 (ATS-Optimized)
Status: Active Development