Professional Experiences

Research Assistant, NSU Intelligent Robotics (NIRO) Lab (June 2025 – Present)

Conducting research on Vision-Language-Action Models (VLAMs) and Reinforcement Learning (RL) for robotics
Developing and testing robotic control strategies using NVIDIA Isaac Sim and Lab for modeling and simulation simulation
Developing and deploying AI models on embedded edge devices (Jetson Orin, Raspberry Pi), integrating perception and decision-making pipelines for real-time robotic applications.
Designing and implementing ROS2-based software systems for robot control, planning, and distributed multi-robot communication.

Assisted in teaching ”Computer Architechture and Organization” course covering topics of MIPS, assembly language, verilog and cpu design
Held office-hours and graded assignments and quizzes

Fashion Trend Research Platform:
- Developed a pipeline to automatically download images from Instagram and tag clothing categories and attributes using multi-modal AI models.
- Developed downstream analytics that converted raw attribute tags into structured insights by aggregating tag frequencies, modeling temporal shifts, and highlighting upward or declining fashion trends.
- Used tools such as OpenAI APIs, deepseek-vl, Nvidia Triton, Docker, PostgreSQL, and Elasticsearch for efficient processing.
Automated Data Extraction from PDFs:
- Created a service to extract specific data from PDFs into JSON format, testing and integrating OCR APIs like Azure Doc Intelligence, Google Cloud Vision API, Amazon Textract, and Tesseract OCR.
- Containerized the service and deployed it on AWS ECS, Fargate, and Lambda for scalability and flexibility.
- Used technologies like Docker, Playwright, and Streamlit to streamline the deployment and user interaction.
Technologies: Nvidia Triton, deepseek-vl, OpenAI APIs, Docker, PostgreSQL, Elasticsearch, FastAPI, Streamlit, AWS ECS, Fargate, Lambda, Azure Doc Intelligence, Google Cloud Vision API, Amazon Textract, Tesseract OCR, instaloader, Playwright.

Implemented the Swin Video Transformer models using TensorFlow
Converted the weights from PyTorch to TensorFlow 2 and uploaded them to TensorFlow Hub
Created notebooks demonstrating how to fine-tune the models

Created dialogue graphs from MultiWOZ dataset using tf-idf, topic modeling, xmeans, dbscan etc.