Swapnil Singh

Curated a dataset by combining data from different data sources like Zillow, US Census, and Eviction Lab and developed SEIFA and gentrification class labels.

Performed a comparative study for classifying and predicting Logistic Regression, Random Forest, Light Gradient Boost Machines, and Naïve Bayes, achieving a classification accuracy range of 85% - 84%.

Employed Seasonal AutoRegressive Integrated Moving Average with eXogenous regressors model to use temporal data of counties in the United States to predict SEIFA scores and early gentrification, giving RMSE score of 4.

Deployed geospatial weighted regressor along with Moron’s I to find the spillover effect in counties and performed cluster analysis to identify the common features in groups of gentrifying clusters.

Proposed policy suggestions for policy makers to control gentrification based on the insights gathered from the study.

Engineered a travel buddy app for 4 user personas using Flutter and Dart for a tailored Marriott experience.

Integrated a matching algorithm with Flask, Agglomerative clustering, and Firebase, boosting match relevance by 80%, and ensuring user privacy with JWT.

Developed a database using MongoDB Atlas of 4063 Virginia Tech webpages by web scrapping using Beautiful Soup and created embeddings using Hugging Face.

Created a Virginia Tech chatbot using RAG, MongoDB Vector Search, and GPT- 4 -Turbo for accurate query responses.

Designed a database with 8 tables using PostgreSQL for managing Imaging, Genetic, and Demographic data for hospitals.

Spearheaded a Flask frontend with RESTful API for 4 personas, adding interactive visualizations using Plotly and Seaborn.

Executed Transformer-based Planning for the Symbolic Regression method using PyTorch to achieve an R2 of 0.74.

Optimized symbolic regression evaluation efficiency by 40% and innovated to reduce computational time by 30%.

Using a custom weight function along with trivial machine learning models for detecting intrusion attacks in UNSW and CIC-IDS 2018 datasets

Designed an algorithm to convert URLs into images and employed a hybrid model of LSTM, GRU, and CNN to classify them into phishing and non-phishing images achieving accuracies in the range of 96%-97% .

Conducted explainable AI analysis on the hybrid model using Grad-CAM and Guided Backpropagation to visualize and interpret the decision-making process for phishing and non-phishing classification.

Leveraged MobileNetV2 for plant disease detection through analysis of leaf images, demonstrating efficiency and innovation in agricultural technology.

Utilized scispacy to identify medical entities in input text and implemented BERT with Named Entity Recognition (NER) for answering questions related to the identified entities, showcasing a robust approach in medical text analysis and question-answering systems.

Leveraged CNNs and image processing techniques of intensity-based segmentation to detect and precisely segment brain tumors in MRI scans

Engineered a versatile system capable of generating five types of questions along with corresponding answers from input text, facilitating output in formats such as Word documents, plain text, or PDF files using GPT2.

Implemented sarcasm detection in conjunction with classification models to effectively identify and filter fake news from benchmark datasets.

Employed a hybrid model combining LSTM, GRU, and CNN architectures to create a chatbot designed to support individuals dealing with mental health challenges, demonstrating a comprehensive and empathetic approach to conversational AI.

Conducted sentiment and emotion analysis on tweets from Indian users discussing the 2016 demonetization in India, providing insights into public reactions during this significant event.

Implemented a Custom CNN for detecting and classifying images in the Hieroglyph dataset, deploying VGG16 as a benchmark for accuracy comparison, demonstrating a thorough evaluation of model performance.

Created a COVID-19 vaccination website akin to Aarogya Setu, incorporating HTML, CSS, JavaScript, PHP, and Machine Learning technologies for a comprehensive and dynamic user experience.

Applied basic machine learning algorithms to classify diverse datasets and meticulously observed their performance, providing valuable insights into algorithm effectiveness across different data domains.

Crafted an informative website for painting auctions in India using HTML, CSS, JavaScript, and PHP, providing a dynamic and engaging platform for art enthusiasts and bidders.

Created an informative website detailing the Prime Ministers of India, employing HTML, CSS, JavaScript, and PHP to offer a dynamic and engaging platform for users seeking comprehensive information.

Designed a straightforward ticketing and tracking system for BEST (Brihanmumbai Electric Supply and Transport) utilizing Tkinter, PyQR, and SQLite libraries in Python, enhancing efficiency and management within the organization.