Phaneendra Babu Gunturu
Machine Learning Research Assistant
Education
Masters in Computer Science, Purdue University, USA, 2025
Bharath Institute of Higher Education & Research, India, 2023
Experience
Deep Learning Researcher
Indiana University Indianapolis, USA
November 2024 - May 2025
Conducting NSF-funded research on large-scale medical datasets using deep learning architectures, including Transformers. Developing and optimizing models for improved accuracy and efficiency, leveraging PyTorch and TensorFlow. Collaborating with a multidisciplinary team to design and implement novel solutions for healthcare applications
Mathematics Tutor
Indiana University Indianapolis, USA
January 2024 - January 2025
Provided one-on-one and group tutoring sessions in calculus, linear algebra, statistics, and probability at the Mathematics DepartmentAssistance Center (MAC), improving student comprehension and academic performance.
Machine Learning Intern
Verzeo, INDIA
June 2021 - Spetember 2021
Led a 5-member team in developing a diabetes prediction system for women, achieving 95% accuracy through ensemble modeling of Random Forest, SVM, and Naive Bayes algorithms while reducing false negatives by 25%. Implemented robust cross-validation techniques and feature selection methods that reduced overfitting by 30%, while optimizing hyperparameters using GridSearchCV to improve model accuracy from 87% to 95%.
Projects
Document Classification System
Architected an enterprise-grade document classification system leveraging Vision Transformers (ViT) and PyTorch, achieving 93.81% accuracy on 400K+ documents with MLflow-based experiment tracking and automated model versioning. Implemented production-ready MLOps pipeline with automated data preprocessing, model checkpointing, and performance monitoring, processing 20K documents/epoch and achieving 98.57% training accuracy. Developed and optimized a multi-label classification model using cutting-edge ViT architecture achieving 95% reduction in document processing time, while maintaining high accuracy across 5 distinct document categories in mortgage domain
Image Captioning System
Developed and implemented an advanced Image Captioning system using TensorFlow and InceptionResNetV2, incorporating attention mechanisms and beam search to generate contextually accurate descriptions with a BLEU-4 score of 0.178. Designed and trained a neural network architecture combining CNN and RNN components, achieving 58.65% word-level prediction accuracy while implementing comprehensive evaluation metrics using NLTK. Created an efficient inference pipeline with beam search algorithm, optimizing caption generation process while maintaining code modularity and error handling for production-ready implementation
Pubg Player analysis & Rank Prediction
Applied advanced data cleaning techniques and optimized dataset memory by 65.5% (983.90 MB to 339.28 MB) while maintaining data integrity and eliminating fraudulent gameplay data. Engineered 15+ game-specific features including normalized kill ratios, movement patterns, and team dynamics to capture comprehensive player performance metrics. Implemented Random Forest Regressor with feature importance analysis, achieving 0.0488 MAE for player ranking prediction while identifying key performance indicators
Restaurant Recommendation System
A full-stack web application for restaurant recommendations with filtering and search capabilities. Features: Restaurant search by city, Price range filtering, Advanced filters for categories and amenities, Restaurant details including ratings and reviews, CRUD operations for restaurants