Rahul Jena

An advanced media pipeline utility that solves file timestamp degradation when downloading bulk photos from Google Takeout. Reconciles decoupled JSON metadata headers with target media files, parsing ISO-8601 creation timestamps, geolocation tags, and camera parameters while safely preserving original file streams.

• Designed core logic to restore original photo timestamps by reconciling Google Photos JSON metadata with local media files.
• Implemented a regex-based file matching and merge algorithm to handle truncated filenames and duplicate naming patterns with high accuracy.
• Ensured non-destructive processing through validation checks and copy-based restoration workflows, with AI assistance used only for boilerplate code.

Engineered a high-precision computer vision pipeline utilizing dlib 68-point facial landmark prediction and Eye Aspect Ratio (EAR) calculations. When prolonged eye closure or micro-sleep indicators exceed the critical threshold, audio-visual alert triggers are dispatched immediately to re-engage the driver.

• Amplified road safety by 15% through the implementation of a real-time drowsiness detection system with a 92% initial baseline accuracy rate.
• Leveraged advanced facial landmark tracking algorithms to achieve 95% accuracy in drowsiness detection, enabling proactive prevention measures.
• Integrated alert systems to provide real-time warnings, enhancing driver awareness and reducing accident risk by 20%.

Designed and built an end-to-end AutoML platform that accepts raw tabular datasets, automatically handles missing value imputation, encodes categorical variables, trains multiple regression algorithms (Linear, Ridge, Random Forest, Gradient Boosting), and ranks them using cross-validated RMSE and R² metrics.

• Strengthened model selection efficiency by developing an AutoML web app, automating data exploration and regression analysis, reducing analysis time by 40%.
• Reduced user friction by 30% through the development of a user-friendly interface for seamless interaction with machine learning models.
• Improved workflow automation by 25%, leading to faster and more accurate model performance evaluation.