I'm a Full Stack Engineer & Solution Architect with 20+ years of experience in mobile development, AI/ML, and cloud solutions.
- Full Stack Development
- AI/ML Solutions
- Cloud Architecture
- Mobile Development (iOS/React Native)
- 🤖 PyTorch
- 🧠 NLP & Transformers
- 👁️ Computer Vision
- ⚡ Generative AI
- RotorQuant – Advanced Quantization Framework for LLM Optimization.
- 🎥 Video Streaming Platform (WWE) World Wrestling Entertainment - Conneticut USA
- 📱 Social Media Platform with AR (8secondz)
- 🚗 Ride-Hailing System - New York City -SOHO (Karhoo)
- 🤖 AI-Powered Solutions
A cutting-edge quantization framework designed to squeeze maximum performance out of LLMs with minimal perplexity loss.
- Weight Transformation: Implements rotational transformations to balance outlier features, making models more "quantization-friendly."
- Hardware Efficiency: Optimized for reduced memory footprints, allowing large-scale models to run on consumer-grade hardware.
- Innovation: Focuses on the intersection of linear algebra and neural compression to mitigate the accuracy drop typical in 4-bit and 8-bit quantization.
- Tech stack: Python, PyTorch, CUDA, Triton.
A recreation of Microsoft's groundbreaking motion synthesis technology
- Implemented neural motion fields for human motion synthesis
- Built using PyTorch and CUDA optimizations
- Focus on performance optimization and memory efficiency
- (update) recently acknowledged here https://github.com/cbsjtu01/IMTalker (see my PR that upgrades code to 58fps)
Implementation study of VASA (Visual Affective Style Transfer) architecture
- Audio-driven facial animation synthesis
- Emotion-preserving style transfer
- Large-scale model training infrastructure
- Tech stack: Python, PyTorch, Audio processing libraries https://wandb.ai/snoozie/vasa-overfitting/runs/8mq2bmob?nw=nwusersnoozie
Reverse engineering study of "MegaPortraits: One-shot Megapixel Neural Head Avatars"
- High-resolution neural head avatar generation
- Implemented novel architecture for one-shot learning
- Utilized Claude AI for code analysis and optimization
- Tech stack: Python, PyTorch, CUDA
Enterprise-grade MLOps toolkit for A100 GPU clusters
- Infrastructure as Code
- Terraform modules for GCP Vertex AI
- Automated cluster provisioning
- GPU quota management
- CI/CD Pipeline
- Docker containerization
- Automated deployment workflows
- Resource optimization
- Monitoring & Scaling
- Custom metrics collection
- Auto-scaling policies
- Cost optimization
ML-powered cryptocurrency trading strategy generator
- Real-time market data processing
- Multiple ML models (Random Forest, LSTM, Transformer)
- Backtesting framework
- Risk management system
- Tech stack: Python, TensorFlow, scikit-learn, pandas
- Advanced computer vision implementations
- Large-scale distributed systems
- MLOps and infrastructure automation
- Performance optimization for GPU clusters
- Research-to-production pipeline development
