Shah Ahmed Masood

Masters in Computer Science student with specialization in AI/ML and cloud computing. Expertise in Python and machine learning, consistently enhancing model performance in collaborative projects. Demonstrated success in mentoring students at AIS, promoting growth in AI research, software Engineering, leadership and teamwork.

AI Skin Health Analyzer Serverless ML Web Application

• Built a production-grade, serverless AI web application for acne severity analysis using PyTorch and AWS Lambda
• Deployed Dockerized ML inference on AWS Lambda, reducing infrastructure costs by 60% com pared to EC2-based deployment
• Integrated a React frontend via AWS Amplify with GitHub-based CI/CD for automated builds and deployments
• Addressed real-world deployment challenges including CORS configuration, payload validation, and Lambda execution constraints, achieving 99%+ successful inference requests in testing

AI Chatbot with AWS Lambda & Bedrock

• Developed a cost-efficient chatbot (50% cheaper than EC2), achieving 99.9% uptime.
• Developed serverless chatbot using AWS Lambda and Bedrock foundation models., Integrated with API Gateway and tested using Postman.
• Monitored logs and exceptions with CloudWatch.

Options Pricing with Deep Learning

• Predicted call option prices with R2 > 0.95 using PyTorch.
• Built a regression-based neural network to approximate European call option prices using Black-Scholes (financial model)., Engineered a synthetic dataset of 100K+ options and standardized features for model training.
• Visualized model performance with predicted vs actual price plots using Matplotlib.

Dog Breed Identification Using Pre-Trained Image Classifier

• Developed an image classification model to identify dog breeds using transfer learning and Computer Vision.
• Achieved high accuracy (over 95%) by fine-tuning model parameters and optimizing the training pipeline.

Fine-Tuning BERT with PEFT

• Worked on Fine-tuning BERT models for optimized performance with minimal computational overhead.
• Used PEFT to optimize NLP performance and implemented tokenization strategies to enhance NLP model efficiency across various datasets.
• Reduced BERT's training time by 35% using parameter-efficient techniques.