🤖 Yuvraj Singh’s AI Portfolio

🎓 Computer Science Engineering Student at IIIT Bhubaneswar (2023-2027)
🔬 Research Focus: NLP, Computer Vision, and Multimodal LLMs
🚀 Mission: Building AI from scratch and bridging research with practical implementation

🌟 What This Portfolio Showcases

This comprehensive portfolio website demonstrates my journey in AI/ML through hands-on implementations, research, and practical applications. Every model, dataset, and experiment represents hours of learning, coding, and understanding the fundamentals of artificial intelligence.

🧠 Core Philosophy

• From Scratch Implementation: Building AI models without relying on pre-built libraries to truly understand the mathematics and algorithms
• Paper-to-Code: Translating cutting-edge research papers into working implementations
• Educational Focus: Creating resources that help others learn AI/ML concepts
• Open Source: All implementations are freely available for the community

🎯 Key Features

🔬 From Scratch AI Models (38+ Implementations)

Location: /models/

Source: Paper-Replications Repository

A comprehensive collection of AI models implemented from scratch, covering:

🤖 Large Language Models & NLP

• Transformer Architecture - The foundation of modern NLP
• BERT - Bidirectional Encoder Representations from Transformers
• GPT - Generative Pre-trained Transformer
• Llama - Meta’s efficient language model architecture
• Llama4 - Advanced multi-expert architecture with MoE
• Kimi-K2 - DeepSeek V3 inspired model with advanced features
• Gemma & Gemma3 - Google’s efficient language models
• DeepSeekV3 - Latest in reasoning capabilities
• Differential Transformer - Novel attention mechanisms
• Attention Mechanisms - Core building blocks of modern AI
• Encoder-Decoder - Seq2seq architectures
• Seq2Seq - Sequence-to-sequence learning
• Fine Tuning using PEFT - Parameter-efficient fine-tuning
• LoRA - Low-Rank Adaptation techniques
• DPO & ORPO - Advanced alignment techniques
• SimplePO - Simplified preference optimization

🎨 Computer Vision & Generative Models

• Vision Transformer (ViT) - Transformers for image classification
• CLIP & CLiP - Vision-language understanding
• CLAP - Contrastive Language-Audio Pre-training
• SigLip - Sigmoid loss for language-image pre-training
• LLaVA - Large Language and Vision Assistant
• PaliGemma - Vision-language model
• Generative Adversarial Networks:
  • DCGANs - Deep Convolutional GANs
  • CGANs - Conditional GANs
  • CycleGANs - Unpaired image-to-image translation
  • WGANs - Wasserstein GANs
  • Pix2Pix - Paired image-to-image translation
• VAE - Variational Autoencoders

🧮 Fundamental Architectures

• RNNs - Recurrent Neural Networks
• LSTM - Long Short-Term Memory networks
• GRU - Gated Recurrent Units
• Mixtral - Mixture of Experts architecture

🎵 Audio & Speech

• Whisper - Speech recognition and transcription
• TTS - Text-to-speech synthesis
• Moonshine - Audio processing models

⚡ Optimization & Training

• DDP - Distributed Data Parallel training

🎮 SmolHub Playground

Location: /smolhub/

Purpose: Experimental AI playground

An interactive space for:

• Proof-of-Concept Models - Testing new ideas quickly
• Educational Experiments - Learning-focused implementations
• Rapid Prototyping - Fast iteration on AI concepts
• Community Contributions - Collaborative learning space

📊 Curated Datasets

Location: /datasets/

Count: 5+ High-quality datasets

Featured Datasets:

1. ImageNet-Mini - 10K images, 100 classes for computer vision
2. SentimentFlow - Advanced sentiment analysis dataset
3. CodeSense - Programming language understanding
4. Anomaly Hunter - Anomaly detection scenarios
5. Multilingual QA - Cross-lingual question answering

Each dataset includes:

• ✅ Preprocessing Scripts - Ready-to-use data pipelines
• ✅ Documentation - Comprehensive usage guides
• ✅ Benchmarks - Performance baselines
• ✅ Visualization Tools - Data exploration utilities

🛠️ Technical Implementation

Backend Architecture

• Framework: Jekyll with GitHub Pages compatibility
• Dynamic Updates: Automated model/dataset synchronization
• Performance: Optimized for fast loading and mobile responsiveness
• SEO: Structured data and meta optimization

AI Model Integration

• Automatic Discovery: GitHub API integration for real-time updates
• Code Highlighting: Syntax highlighting for multiple programming languages
• Documentation: Auto-generated documentation from README files
• Version Control: Git-based versioning for all implementations

Deployment

• Platform: Render.com with automatic deployments
• CI/CD: GitHub Actions for continuous integration
• ZeroGPU: Hf ZeroGPU support for hosting my models as spaces!

📈 Impact & Learning Outcomes

Educational Value

• 📚 Deep Understanding: Every model built from mathematical foundations
• 🔍 Research Translation: Converting papers into working code
• 🎯 Practical Application: Real-world implementation experience
• 🤝 Community Learning: Open-source contributions for others

Technical Skills Demonstrated

• Programming: Python, PyTorch, TensorFlow, JavaScript, Ruby
• Mathematics: Linear Algebra, Calculus, Statistics, Probability
• Algorithms: Deep Learning, Machine Learning, Computer Vision, NLP
• Software Engineering: Version Control, Testing, Documentation, Deployment

Research Engagement

• 📑 Paper Implementation: 38+ research papers translated to code
• 🔬 Experimental Design: Systematic approach to model development
• 📊 Performance Analysis: Benchmarking and optimization
• 📝 Documentation: Comprehensive technical writing

🚀 Getting Started

Explore the Portfolio

1. Visit the Website: Yuvraj’s Portfolio
2. Browse Models: Check out /models/ for from-scratch implementations
3. Try SmolHub: Experiment in /smolhub/ playground
4. Download Datasets: Access curated data in /datasets/

For Developers

# Clone the repository
git clone https://github.com/YuvrajSingh-mist/yuvraj-singh-portfolio.github.io.git

# Install dependencies
bundle install

# Run locally
bundle exec jekyll serve

# Visit localhost:4000

For AI Researchers

• 🔗 Source Code: Paper-Replications Repository
• 📖 Documentation: Each model includes comprehensive README
• 🤝 Collaboration: Issues and pull requests welcome
• 📧 Contact: Reach out for research discussions

🎯 Current Focus & Future Goals

Current Work

• 🔬 Fine-tuning LLMs - Advanced optimization techniques
• 📚 GAN Research - Exploring generative adversarial networks
• 🤖 Multimodal Models - Vision-language understanding
• 📊 Model Optimization - Efficiency and performance improvements

Seeking Opportunities

• 🎯 Research Internships - NLP and Computer Vision roles
• 💼 Full-time Positions - RE/RS roles in AI/ML
• 🤝 Collaborations - Open to research partnerships
• 🎓 Mentoring - Helping others start their AI journey

🤝 Connect & Collaborate

### Professional Links

• Email: yuvraj.mist@gmail.com
• X: https://x.com/YuvrajS9886

  Contributing

  This portfolio is open-source and welcomes contributions:

• 🐛 Bug Reports: Issues and suggestions
• 💡 Feature Requests: New ideas and improvements
• 🤝 Code Contributions: Pull requests welcome
• 📚 Documentation: Help improve explanations

🏆 Recognition & Stats

• 🤖 38+ AI Models implemented from scratch
• 📊 5+ Datasets curated and documented
• 🌟 Open Source - All code freely available
• 🎓 Educational Impact - Helping others learn AI
• 🚀 Active Development - Continuously updated

📜 License

This portfolio and associated code are released under the MIT License. Feel free to use, modify, and distribute with appropriate attribution.