Devank Shinde

Adept in leveraging Python and modern machine learning frameworks to build intelligent, data-driven applications. Demonstrated experience in developing NLP models, federated learning systems, and AI-enabled solutions, supported by a strong foundation in data analysis, and software engineering.

Multilingual Sentiment Analysis

Conducted data preprocessing and developed a testing dataset using mixed-region Marathi news articles from Divya Marathi, followed by sentiment labeling through translation workflows and pretrained NLTK sentiment models, performed training of multiple machine-learning models for Hindi and Marathi sentiment analysis, selected appropriate evaluation metrics, and carried out comparative analysis. Additionally, implemented a 5-fold cross-validation approach to ensure model robustness and reliable performance evaluation for Marathi language sentiment classification

https://github.com/devankshinde1/SentimentAnalysis 

MENTOR

Contributed to the design and implementation of the recommendation component by integrating adaptive feedback and context-specific learning resource suggestions based on learner performance, involved analyzing response patterns, aligning resources to question topics and difficulty levels, and enhancing the system's personalized learning workflow within the MENTOR platform

https://github.com/devankshinde1/MENTOR 

Federated Learning for Smart Healthcare

Led the end-to-end development of the federated learning model by designing a multi-task CNN architecture, implementing client-side preprocessing and training pipelines, and optimizing federated aggregation to enhance model accuracy and AUC, coordinated team efforts during model building, and integrated blockchain-based access control to ensure secure and privacy-preserving predictions across medical imaging modalities

https://github.com/devankshinde1/FL

Research Paper:

Enhancing Privacy and Personalization in Federated Learning, IJNRD, Volume 10, Issue 4, April 2025, 

Published on 4 April 2025. Led end-to-end development of a federated learning model using a multi-task CNN, implementing client-side preprocessing, and optimizing aggregation for improved accuracy and AUC Coordinated team contributions and integrated blockchain-based access control for secure, privacy-preserving medical image predictions.

Sentiment Analysis for Hindi and Marathi Languages using RoBERTa Model, Registered Copyright (Computer Software Work), Copyright Office Government of India, Registration No. SW-18425/2024, dated 13 March 2024. The work focuses on enhancing sentiment analysis for Hindi and Marathi languages using the RoBERTa language model and Python.