SAOUD YAHYA

A modern e-commerce platform where microservices work in harmony to deliver exceptional shopping experiences. Features event-driven architecture, service mesh, and comprehensive observability.

• Built 8 microservices using Java Spring Boot and Go, deployed on AWS EKS
• Complete GitOps automation with ArgoCD for zero-downtime deployments
• Event-driven architecture with service mesh (Istio) and async Kafka communication
• Seamless kind-to-EKS production migration with one-command deployment
• Comprehensive observability: Prometheus, Grafana, ELK Stack
• Multi-environment CI/CD with customer frontend (Next.js) and admin dashboard

Developed a comprehensive Kubernetes project with three Flask applications: LSTM for fake news detection, an ML pipeline for S&P 500 prediction, and GPT-2 for text generation. Deployed and managed using kubectl with comprehensive monitoring through Prometheus and Grafana for real-time insights into scalable and reliable ML workflows.

• Three specialized Flask applications: fake news detection, S&P 500 prediction, text generation
• LSTM neural networks for sequential data processing and prediction tasks
• GPT-2 integration for advanced natural language generation capabilities
• Kubernetes deployment with kubectl for container orchestration and scaling
• Real-time monitoring and metrics collection using Prometheus and Grafana
• Scalable ML workflows with automatic resource management and fault tolerance

Developed an innovative Named Entity Recognition (NER) system for extracting animal entities from scientific PDF documents using self-supervised learning techniques. This project addresses the critical challenge of traditional NER systems that require massive manually annotated datasets, which typically cost $50,000-$100,000 per domain and take 6-12 months for expert annotation.

• Self-Supervised Entity Relationship Learning eliminating expensive manual annotation
• Ultra-Strict Filtering mechanisms ensuring high-quality entity extraction
• Contrastive Learning methods improving recognition accuracy and relationships
• Cost-effective alternative to domain-specific datasets with cross-discipline generalization
• Scalable solution for biodiversity research and scientific literature analysis
• Overcame traditional NER limitations and computational expense of standard transformers

Conducted comprehensive research investigating scaling behaviors of standard Transformers vs Switch Transformers, examining how model parameters affect performance and internal representations. This work addresses fundamental questions about optimal model design in large-scale neural networks.

• Architecture-Dependent Scaling Discovery with opposite coefficients (-0.0381 vs +0.0173)
• Enhanced Mechanistic Interpretability Framework using SAE-enhanced superposition analysis
• Performance Optimization with Switch-Large achieving best performance (120.91 PPL)
• Analyzed 16 model configurations ranging from 13.1M to 597.3M parameters
• Developed superposition scoring methodology with consistent sparsity patterns
• Created practical guidelines for architecture selection in resource-constrained environments

Intelligent accessibility solution combining computer vision and NLP to assist visually impaired users with real-time environmental awareness and voice interaction. Winner of accessibility innovation awards.

• Real-time object detection and scene description using YOLOv9
• Voice interaction with AI-powered natural language processing
• Hardware-software integration with mobile companion app
• Containerized deployment for scalability and maintainability

📸 Project Showcase

Smart Glasses Sketch

Smart Glasses Hardware

Mobile Companion App

Sophisticated LLM-based content generation system with retrieval-augmented generation capabilities using LangChain and multi-agent orchestration for intelligent decision-making.

• Multi-agent RAG systems with intelligent orchestration
• Real-time generation with vector database integration
• Horizontal scaling with containerized deployment
• Advanced prompt engineering and context management

Scalable microservices ecosystem built with Spring Boot, Spring Cloud, and Docker, featuring service discovery, distributed tracing, and comprehensive CI/CD pipelines.

• Microservices with service discovery and load balancing
• Asynchronous messaging with RabbitMQ integration
• Distributed tracing with Zipkin for monitoring
• End-to-end CI/CD pipelines with Jenkins automation
• Multi-database support (PostgreSQL & MongoDB)