ROS2 Object Tracking and Control System

Built a real-time ROS2-based object tracking and control system integrating computer vision, state estimation, and closed-loop robotic control.

Implemented object detection and tracking using OpenCV, stabilized motion estimates using a Kalman Filter, and generated smooth velocity commands via PID control.

Designed a modular ROS2 pipeline with camera, perception, tracking, and control nodes, ensuring robust real-time performance under noisy and intermittent detections.

Predictive Maintenance – RUL Estimation (LSTM)

Developed a deep learning–based predictive maintenance system to estimate Remaining Useful Life (RUL) of industrial machinery using multivariate sensor time-series data.

Designed an end-to-end pipeline with data normalization, sliding-window sequence generation, and LSTM-based regression to capture long-term degradation trends.

Evaluated models using RMSE and MAE, demonstrating improved failure prediction accuracy over traditional threshold-based maintenance approaches.

Industrial Surface Defect Detection (Computer Vision)

Built a deep learning–based computer vision system to automatically detect industrial surface defects such as cracks, scratches, and dents from high-resolution images.

Implemented CNN and transfer-learning pipelines using PyTorch, with extensive data augmentation and class-imbalance handling for robustness under varying lighting and surface conditions.

Optimized performance using precision, recall, and F1-score to minimize false negatives, demonstrating applicability to real-world industrial inspection workflows.

AI Agent for Operating Backend System

Built an AI agent using Gemini-pro LLM, automating workflows with 50% improved efficiency and 80% fewer manual tasks at 95% accuracy.

Developed a Flask API with SQLite, cutting response times by 40%, supporting 1,000 queries, and boosting backend efficiency by 30%.

CardioCare

CardioCare is a heart disease management system that leverages machine learning to predict and assess cardiovascular risks based on real-time patient data.

The platform provides healthcare professionals with valuable insights, enabling proactive and informed decision-making for better patient care.

Click below link for Live Website.

Air Calligraphy Using Computer Vision

Designed a computer vision-based system enabling individuals without limbs (hand) to write in the air using a fingertip, eliminating the need for traditional input devices.

Leveraged OpenCV and Python to implement real-time object detection and gesture recognition for intuitive and accessible air-writing.

NYC Crime Analytics - using Preswald

The NY Crime Analytics Application is an interactive dashboard built using Preswald, providing a comprehensive view of crime trends across New York State from 1990 to 2023.

The application allows users to explore county-wise crime statistics, including violent, property, and firearm-related crimes, with visually engaging charts and interactive filtering for deeper insights.

Monument Intelligence Dashboard

The Monument Intelligence Dashboard is an interactive, AI-powered web app built with Preswald, allowing users to explore and analyze iconic monuments worldwide through dynamic visualizations and natural language queries.

The application features an interactive geospatial map, trend analysis by country and century, and a chatbot for intuitive, conversational exploration of global monuments' historical data and visitor statistics.

BlinkIt Sales Analytics

Developed an interactive Power BI dashboard to analyze Blinkit's sales data, providing insights into key metrics such as total sales, item distribution, outlet performance, and customer preferences.

Implemented dynamic filtering and visualizations to track sales trends by product type, outlet size, and location, enabling data-driven decision-making for business growth.