White Turtle Studios

Utilized and fine-tuned AI models for voice generation, RVCs, subtitles, text-to-video using Runway, and image creation using open-source stable diffusion models, creating over 200 media files. Demonstrated rapid adaptability by implementing novel models and crafting optimal prompts, enhancing media output realism and quality. Collaborated with the creative team to integrate AI-driven solutions, boosting workflow efficiency and output quality. Researched about such AI applications, maintaining a leading edge in emerging technologies.

Node.js Backend

During the freelancing tenure, specialized in developing robust backend infrastructures using Node.js. Created and maintained over 50 RESTful APIs for clients such as Blue Lotus and Zetta, ensuring seamless integration and high performance. Leveraged Express.js for server-side functionality, optimized database interactions, and implemented efficient middleware solutions. Additionally, provided guidance to clients on best practices for scalable architecture, enhancing the reliability and efficiency of their applications. This hands-on experience solidified expertise in Node.js, backend development, and client collaboration.

DR-SASV: A DEEP AND RELIABLE SPOOF AWARE SPEECH VERIFICATION SYSTEM

International Journal of Information Technologies and Security

The DR-SASV system integrates a CNN-based spoof detection model with a transformer-based speech verification model to identify both impostor speakers and spoofing attacks. Trained on the ASVSpoof 2019 LA dataset, it achieves 96% accuracy in spoof detection and a 13.74% error rate in speech verification, with an overall equal error rate of 10.32%. This end-to-end solution offers robust protection against fraud.

STOCK PREDICTION USING GAN AND SENTIMENT ANALYSIS

International Journal on Recent and Innovation Trends in Computing and Communication

This paper proposes a method for forecasting stock market movements based on sentiment analysis of tweets. By mapping Twitter data to stock prices via the Twitter API, a dataset is created to assess public perception. The method employs a Generative Adversarial Networks (GAN) framework, integrating three GAN models with Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM). The Root Mean Square Error (RMSE) metric evaluates the model's accuracy, determining which GAN model provides the best results.

SMARTPHONE-BASED POTHOLE AND SPEED-BREAKER DETECTION SYSTEM

Part of the book series: Smart Innovation, Systems and Technologies (Springer,SIST,volume 376)

Potholes are a significant road hazard, causing vehicle damage and endangering drivers. Traditional manual inspection methods are slow and costly. This research introduces a technical approach using accelerometer data and statistical methods like CUSUM filtering and mean value filtering for automated pothole detection. Accelerometer data, collected inexpensively, undergoes preprocessing with these methods to detect indicators such as sudden acceleration changes and vibrations associated with potholes. The algorithm is trained and tested on both secondary and primary datasets created specifically for this purpose, using labeled data to detect potholes accurately. Experimental results during training and testing on primary data show that the SVM is effective at detecting potholes with an accuracy of 93.3%.

OFFENSE DETECTION USING BERT AND CNN

IEEE Xplore - 2022 IEEE 3rd Global Conference for Advancement in Technology

In today's world of abundant user-generated content (e.g., articles, books, blogs), concerns about offensive material are growing. Unlike RNN and LSTM, which process words sequentially, BERT analyzes input simultaneously. This study combines BERT with CNN layers to enhance accuracy in classifying hate speech and offensive language. Initially trained on a dataset with three categories (hate, offensive, neither), focusing later on offensive and non-offensive categories improved BERT's accuracy. Without CNN, BERT achieved 85% accuracy, whereas with CNN focusing on two categories, it reached nearly 95%

Indian Phishing Landscape- ML & DL approach for detecting malicious URLs

Awaiting review - This study categorizes phishing websites using ML algorithms and employs CNN, RNN, and LSTM for effective detection. It introduces a curated Indian Phishing Website Dataset tailored specifically for URL detection in the context of cybersecurity.

SwasthyaBuddy: Assessing Chronic Obstructive Pulmonary Disorder Patients

Awaiting review - This research highlights the critical role of Machine Learning and Deep Learning Algorithms in assessing Chronic Obstructive Pulmonary Disorder patients, emphasizing the development of a system aimed at improving diagnosis and patient care outcomes.

Unearthing Code Smells: In-Depth Exploration of ML in Code Smell Detection

Awaiting review - This study provides a comprehensive review of various types of code smells and evaluates the efficacy of ML algorithms in detecting and mitigating them, contributing to the advancement of software evolution.

Decoding Plasmodium knowlesi Stages with Explainable AI and ML/DL Models

Leveraging ML/DL models and Explainable AI, this study advances malaria diagnostics by accurately detecting and understanding Plasmodium knowlesi stages through single-cell RNA sequencing, ensuring precise and insightful results.

3rd International Conference on Advanced Computing Technologies and Applications, Logistics Department 2023

Contributed to event decoration, management for judges, participants, and keynote speakers, ensuring a seamless experience.

Teaching Assistance, Dwarkdas J Sanghvi College of Engineering 2023

Created comprehensive lab manuals, code sets, and assignments for Image Processing & Computer Vision. Developed hands-on experiments in YOLO, GANs, transfer learning, and motion analysis.

Volunteer, COVID-19 Vaccination Drive

Assisted in organizing and supporting a successful vaccination drive in a village near Mumbai, ensuring smooth operations and community participation.