My Projects
Exploring the intersection of AI, Machine Learning, and real-world applications
Projects
Intelli-Chat
AI-Powered Conversational System
Deployed an AI chatbot on GCP processing 60,000+ documents with hybrid retrieval-augmented generation approach.
Multi-Language Translation System
Neural Machine Translation
Built production-ready neural translation system supporting 5+ language pairs with 85% BLEU score using sequence-to-sequence architecture.
Tokenization Challenges in Multilingual GPT
NLP Research & Implementation
Researched and implemented optimized tokenization for multilingual language models, improving efficiency by 60% for non-English languages.
Meme Persuasion Detection
Multimodal Content Analysis
Developed multimodal classification system for detecting persuasion techniques in internet memes with 78% accuracy using BERT and ResNet-50.
Research Projects
Published academic research in AI and machine learning
Tokenization Challenges in Multilingual GPT
Guide: Prof. Rohini Srihari
Optimized tokenization in multilingual language models, improving efficiency by 60% for non-English languages. Developed language-specific preprocessing pipelines, reducing token usage per prompt from 70-100 to 18-25. Conducted comparative analysis between native script and transliterated text, demonstrating significant computational advantages. Created a specialized preprocessing framework for Telugu language that can be extended to other Indic languages, enhancing the accessibility of large language models for non-English users.
Fraud Detection in Automobile Insurance Claims using Machine Learning Algorithms
Guide: Prof. Nasrin Akhter
Engineered ML pipeline for fraud detection achieving 78% accuracy and 81% AUC using ensemble methods. Implemented data balancing and feature engineering techniques to optimize model performance on imbalanced datasets. Developed ML models (RF, KNN, DT, SVM) for fraud detection, selecting Random Forest as optimal. Conducted extensive evaluation despite data imbalance, demonstrating RF's superior performance over other models.
A Quantitative Analysis of Basic vs. Deep Learning-based Image Data Augmentation Techniques
Guide: Prof. Rohini Srihari
Conducted research on deep learning-based image augmentation techniques, achieving 98.57% accuracy on MNIST dataset. Implemented and evaluated various data augmentation strategies to improve model robustness and generalization. Evaluated computational burden, storage requirements, and performance to recommend cost-effective augmentation strategies. Published findings in IEEE ICSES 2021, contributing to enhanced robustness in deep learning pipelines.
Interested in Collaboration?
I'm always open to discussing new projects, creative ideas, or opportunities to be part of your vision.
Get in Touch