Tushar Budhwani

Optimizing Self-written Machine Learning Algorithms for Real-World Data Applications

May 2024

• Conducted a comparative analysis of self-written models (Neural Networks, Random Forests, and KNN) on 4 diverse datasets.
• Designed and implemented multi-layer architectures with custom back propagation, activation functions (ReLU, Sigmoid), and weight initialization, achieving up to 98.14% test accuracy on the Handwritten Digits dataset.
• Executed comprehensive k-fold cross-validation on all models, ensuring robust generalization, reduced overfitting, and consistently high accuracy and F1 scores.

Personalizing LLMs (Large Language Models) based on User Profile

Dec 2023

• Created a Retrieval-Augmented Generation (RAG) pipeline to personalize LLM outputs by integrating user profile data.
• Engineered enhanced prompts by combining input queries, retrieved articles, and lexical augmentation techniques, improving the LLM's contextual understanding and alignment with user preferences by 14%.
• Implemented a BM25 retriever to extract top-k (1,2,3) relevant articles from a user's profile.
• Fine-tuned Google's FLAN-T5 model to predict article categories, boosting accuracy by 23%.

ML-Based Web Platform for Disease Detection

Feb 2023

• Developed and evaluated multiple ML models (SVM, Random Forest, XGBoost, AdaBoost), using hyperparameter tuning and k-fold cross-validation to identify the best-performing model for each disease.
• Implemented preprocessing techniques such as feature engineering, SMOTE, and standardization, improving robustness and boosting prediction accuracy of the models by up to 4%.
• Constructed a Stacking Ensemble model (Logistic Regression, KNN, Decision Trees) with a leading 91.3% accuracy in heart disease prediction among seven ML models.
• Assembled a Stacking Ensemble model combining Logistic Regression, KNN, and Decision Trees, attaining 91.3% accuracy for heart disease detection, outperforming individual models.

Intelligent Food Management Application

Nov 2022

• Engineered a full-stack web app using Laravel(PHP) with individual modules for expiration tracking, recipes, inventory management, nutrition optimization, and notifications for items nearing expiration.
• Created a responsive and visually engaging front-end interface using HTML5, CSS3, Bootstrap, and Sass, ensuring cross-device compatibility and an intuitive user experience.
• Implemented a smart recipe recommendation engine using keyword-based matching against a API-based recipe dataset.

Social Media platform for Researchers and Developers

Mar 2022

• Developed a web application for programmers and researchers to efficiently collaborate and network.
• Engineered a recommendation system connecting users with relevant projects and peers, leveraging skillsets and coding styles.
• Formulated a data-driven user matching algorithm using weighted scoring to connect researchers and developers based on shared interests and skills, enhancing user experience and interdisciplinary collaboration.
• Enhanced database performance by optimizing SQL queries and introducing indexed views in MySQL, which sped up data retrieval by 40% for complex searches.
• Achieved Top 10 ranking among 120+ competing teams in a Hackathon, showcasing exceptional project execution and innovation.

Technical Head at National Service Scheme (NSS) - TSEC, University of Mumbai

Sep 2021 - Jun 2022

- Organized over 80 community service events including Blood Donation Drives, Cleanup Drives, etc.