Rohan Mecheri

Experience at Pando, Indium, Crayon Data, and CoderDojo

• BayesTune

• - Developed a distributed hyperparameter tuning system leveraging Bayesian Optimization and multiprocessing for parallel model evaluation.
• - Automated hyperparameter search for SVM models with dynamic result logging and Dockerized deployment.
• - Achieved 97.2% cross-validation accuracy in hyperparameter tuning for SVM models, with optimally computed hyperparameters

• Predictify

• - Developed a predictive maintenance framework using Keras and LSTM neural networks to estimate Remaining Useful Life (RUL) of industrial equipment.
• - Implemented advanced feature engineering techniques, such as time-series decomposition and lag features, to extract insights from sensor data.
• - Conducted rigorous hyperparameter tuning, achieving a 7% improvement in Mean Absolute Error (MAE) compared to a baseline model.

• PhishEye

• - Developed a phishing detection system using Natural Language Processing (NLP) techniques, including Logistic Regression and Support Vector Machine classifiers, to identify malicious URLs based on lexical patterns and sentiment analysis.
• - Achieved an 89% accuracy rate by utilizing custom vectorization techniques.

• BetterJava, a Java pre-processor that translates code written with indentation into brackets

• Chennai Volunteers, a micro-volunteering portal for the Giving Matters Foundation.

• Instructor

• - Supplemental Instructor for Carnegie Mellon's 21-260 - Differential Equations.