Projects
These projects encompass various areas of computer science, including machine learning, natural language processing, data visualization, and software security. They highlight the practical applications and theoretical knowledge I have acquired during my studies. Please click on the heading of each project to view it completely.
Linguistics for Indian Languages
Adopted Unicode correction on a mother tongue corpus and analyzed unigram and bigram frequencies using BPE, mBERT, indicBERT, and White-space tokenizers to identify word groups and syllables. Fine-tuned IndicBERT and IndicNER models for named entity recognition and evaluated machine translation models NLLB-200, IndicBART, and ChatGPT for English-Indian language translations.
Audio Aura
Conducted a comprehensive analysis of global music trends, utilizing advanced data visualization techniques to uncover insights on genre preferences, artist popularity, and emerging musical trends. Deployed a KNN-based music recommendation system using attributes like Danceability, Energy, Valence, and Tempo to suggest songs, supporting dynamic SQL querying for refining recommendations across 67 countries.
Linear CARPUF
Proposed a Linear CARPUF model leveraging machine learning to enhance cryptographic security by simulating unique, unclonable device-specific "handshakes." Refined logistic regression algorithms by incorporating the Khatri-Rao product, achieving a 30% increase in operational efficiency and enabling real-time generation of challenge-response pairs for complex scenarios.
Chiron Framework: Kachua
Applied symbolic execution and the Z3 theorem prover for program synthesis, identifying constant assignments to establish semantic equivalence between two programs with constraints. Examined a Spectrum-Based Fault Localization technique using Density Diversity Uniqueness and Suspiciousness with Ochiai metric to detect and localize program bugs effectively.
Edu-360: Academic & Innovation Hub
Implemented key features like Learn Space for study resources, Exam Space for past year questions, and a Blog Section for user-generated content. These features were built using PHP, HTML, CSS, and Bootstrap, and integrated with SAWO Labs API for enhanced functionality.
Fact Filter using NLP
Developed a machine learning model to filter facts and detect false news stories. The model used text feature extraction techniques like TF-IDF, Word2Vec, Regex, and Bag of Words to differentiate between authentic and fraudulent news stories.
UDAAN Tool Analysis
Analyzed UDAAN, a post-editing tool developed by IIT Bombay for document translation using Optical Character Recognition. The analysis demonstrated that UDAAN achieved a 3.5 times increase in translation speed compared to starting from scratch and post-editing over Machine Translation.
Phishing Website Detection
Gathered 17 features from URLs, covering attributes like address bar information, domain characteristics, and HTML/JavaScript analysis to improve detection accuracy. Executed models such as Decision Tree with 81.4% accuracy, Random Forest with 81.8%, XGBoost achieving 86.6%, and Autoencoder Neural Network with 16.1% accuracy.
