Designed and provisioned AWS infrastructure with Terraform for a production grade web application deployment.
AWS | Terraform | Zero TrustDesigned and implemented a microservices-based DevOps architecture for an e-commerce application using Docker, Kubernetes, and AWS EKS.
DevOps | Kubernetes | EKSBuilt a scalable serverless receipt processing system on AWS using Terraform, Lambda, S3, API Gateway, and DynamoDB to support multi-user workloads.
Serverless | OCR | AWSDesigned and provisioned AWS infrastructure using Terraform to deploy EC2 instances, security groups, and IAM roles following least privilege principles.
GenAI | Bedrock | AWSBuilt a highly available and scalable WordPress architecture using multi-AZ EC2, Auto Scaling, and ALB with LAMP and LEMP stack configurations.
AWS | WordPress | HADeveloped an event-driven pipeline that processes text files uploaded to Amazon S3, performs word frequency analysis, and stores results in DynamoDB.
AWS | Step Functions | LambdaDeveloped and published a Streamlit web application for classifying English accents from YouTube video links.
Deep Learning | Web AppCreated a fully reproducible ML workflow for model versioning, experiment tracking, and deployment.
MLOps | Data VersioningDesigned a deep neural network leveraging minimal scribble-based user input and convexity constraints for robust region delineation.
Deep Learning | Computer VisionDeveloped an inverse kinematics model for a 3 DOF manipulator robot to execute precise pick, place, and homing operations.
Robotics | Control SystemsHands-on mastery of foundational LLM concepts, including tokenizer development, architecture coding, and fine-tuning models like GPT-2 and Llama 2.
LLMs | Deep LearningSelected deep learning, robustness, and systems work grouped into a single expandable project card.
Trained a supervised CNN (TensorFlow Keras) on a Simpsons character dataset (90% accuracy) and used Transfer Learning (PyTorch) on a custom dataset (89% accuracy).
Integrated ResNet-18 and ConvNeXt-Tiny into the AugMix framework to evaluate model robustness under distribution shifts, tracking results with TensorBoard. Code
Modified a CLI tool to sort BMP image columns by average RGB value using C++17 and replicated the logic in Python. Used CMake for builds.