Prashanna Raj Pandit

Research in gait analysis and AI-based motion understanding using deep learning and signal processing.

• Designed and implemented end-to-end data processing pipelines by extracting 2D pose keypoints from noisy video datasets using OpenPose; processed large-scale JSON data with Python (Pandas, NumPy), OpenCV, and Bash scripting to compute clinically relevant gait features such as gait cycle, stride length, cadence, and joint angles.
• Applied statistical methods and feature engineering to gait time-series data, including cubic spline interpolation, Savitzky–Golay smoothing, gait phase segmentation, and turn removal; constructed a clean, multimodal dataset optimized for deep learning models and visualized gait patterns using Matplotlib and Seaborn.
• Developed and trained sequential deep learning models (LSTM) for autism classification, achieving 88% accuracy, and performed model optimization through validation, class balancing, and performance analysis.
• Packaged trained machine learning models using Docker and deployed them through RESTful APIs, integrating them into a production-ready inference pipeline with CI/CD automation to support scalable, repeatable deployments and seamless integration with a web-based application.

Full-stack web development and backend optimization for scalable AI-ready systems.

• Built and deployed a Flask-based web application with RESTful APIs, leveraging SQL-backed structured data models and JSON-based data exchange to support analytics, reporting, and backend business logic; improved system performance and SEO metrics by 20%.
• Designed and maintained data workflows using SQL queries for data retrieval, transformation, and validation, enabling reliable storage, querying, and integration of application data across services.
• Implemented CI/CD pipelines using GitHub Actions and Docker in a Linux environment, automating testing and deployment and reducing manual release effort by 40%, while collaborating with the team using Git and Bash scripting.