• Developed a comprehensive full-stack AI application using Next.js, ensuring smooth project delivery (DevOps) through a streamlined CI/CD pipeline with GitHub Actions, Docker, and Kubernetes (K8n) on Azure. • Build an AI agent to talk to a SQL database in natural language for CRUD operations using LLMs as function blocks.• Developed a secure microservice architecture for a finetuned Open-Source Large Language Model LLM using PyTorch to work on sensitive medical data. Ensured data security through encrypted pipelines.• Implemented advanced Retrieval-Augmented Generation methods, enhancing embedding retrieval accuracy to 94%. • Authored extensive testing using PyTest, covering unit tests, integration tests (Stripe API), validation of FastAPI endpoints, and CRUD operations to maintain high code quality and ensure robust system functionality. • Created AI-driven agentic systems using LLM from MixtralAI and OpenAI, enhanced with CrewAI/Autogen for automating business processes, including customer support, appointment booking, and lead qualification.

• Achieved 80% tracking accuracy on the MOT20 dataset using advanced Vision Transformers using PyTorch. • Enhanced object tracking and detection by reducing ID switching errors by 15% using a mix of deep learning models like CNN, YOLO, and VIT (vision transformer)• Optimized the neural network, achieving a 30% reduction in size through efficient pruning and knowledge distillation. • Streamlined and containerized key ROS2 modules, including Nvidia’s Nvblox and V-SLAM, for improved target detection and localization while maintaining code using CI/CD pipeline (DevOps).

Artificial intelligence I (CS411) Teaching Assistant at University Of Illinois at Chicago

• Implemented real-time obstacle detection & avoidance using localizations using RGBD scans and hybrid A*, written in Python, using PyTorch (CNN and YOLO) and OpenCV. • Tested mission planning algorithms and solutions on Gazebo sim using ROS 2 as middleware. • Performed computation for AI on edge using Jetson Nano and Jetson SDK.

Robotics/Machine learning research•Generated Synthetic labeled dataset using unity and Nvidia Issac; And trained a custom object detection model with DetectNetv2 as backbone.•Generated 3D Object Pose Estimation of the Object of interest by using Pose CNN Decoder provided by Jetson SDK.•Used stereo depth sensing camera to create the map of the surrounding for Navigating and Interacting in 3D space (SLAM).

• Used Python, beautiful soup, and SQLite for web scraping data from multiple resources of interest• Collected and stored data on NVIVO for research on reliability and validity of various educational products• Focused on effectively cleaning, labeling, and providing insights on the measure data (Educational products)

• Reduced IOT automation programming time by ~200% by implementing the k-means cluster on user interaction data, using Machine Learning libraries like Keras(Tensorflow backend), SkLearn, Pandas, and Numpy. • Improved code base by implementing event-driven callbacks using Flask for fast communication between IOT devices. • Recognized by the CEO of the company for execution and timely completion of projects worth $6.7M.