Kunal Shah Email & Phone Number

Chicago, Il, Us

Raleigh, North Carolina, Us

Currently working on Leveraging LLMs in enhancing both feature extraction from unstructured data and user interaction.- Feature Extraction: Extract insights from unstructured data such as commit messages or issues using GPT-4.- User Interaction: Enable natural language querying and responses for user-friendly interfaces using ChatGPT.- Code and Issue Labeling: Categorize and evaluate the impact of code commits and issues using CodeBERT.

Raleigh, North Carolina, Us

Machine Learning Expertise for Open-Source Computational Science (OSCS) Health Prediction:- Developed a transferable prediction model (95% accuracy) using hierarchical BIRCH clustering with Random Forest Regressor, to assess OSCS project health, analyzing 150+ GitHub projects.- Conducted statistical tests to assess the accuracy of the predictive models.- Leveraged SHAP and LIME for interpretable insights, contributing to project sustainability and stakeholder trust.- Automated data mining from diverse sources (150+ GitHub projects) for large-scale health assessment.Optimized Regression Models for OSCS Health Metrics:- Implemented HyperOpt to optimize regression models for predicting OSCS health metrics, reducing mean squared error by 40%.- Improved model performance and efficiency for more accurate and timely health assessments.- Evaluated and compared various regression techniques for optimal model selection.- Conducted extensive literature review and documentation as a part of the research.Open-Source Contribution and Research Collaboration in Computational Science:- Developed and shared open-source tools utilizing data mining and CART-based models for OSCS health prediction, contributing to the field's advancement.- Participated in an NSF-funded research project on adapting Empirical SE to Computational Science, applying expertise in hierarchical clustering and ensemble learning(NSF Award: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1931425).

Boston, Ma, Us

Developed and Deployed Generative AI Model:- Built a production-ready text-to-image model using AWS SageMaker for Fidelity's internal platform.- Leveraged Prompt Engineering and LoRA to reduce errors and accelerate fine-tuning for specific use cases (portraits, digital art, Adobe Firefly).Implemented Comprehensive Data Analysis Pipeline:- Evaluated a potential Fidelity software for automated data analysis on a massive 2.1 million-participant dataset.- Performed rigorous Exploratory Data Analysis (EDA) including data cleaning, transformation, feature engineering, and advanced data visualization. Created comprehensive dashboards using AWS SageMaker and TensorBoard to monitor metrics, detect data leakage, analyze data correlation, and present statistical plots. - These visualizations enabled deeper insights into data patterns and facilitated the presentation of evaluations and recommendations to senior management, significantly impacting decisions on software adoption.

New York, Ny, Us

Designed and deployed a user-driven NLP data solution:- Reduced information gathering time by 70% compared to manual methods.- Empowered users to define specific topics, interests, and timeframes for personalized information retrieval.Developed a high-accuracy web scraping tool (96%):- Leveraged popular Python libraries (Beautiful Soup, Scrapy, Selenium) for scraping and data extraction.- Employed deep learning models (LSTM, BiLSTM, Transformers) for document summarization, sentiment analysis, and natural language understanding.- Achieved 96% accuracy in data extraction from diverse sources (webpages, images, databases, social media).Seamless deployment and user interaction:- Deployed the tool on the Google Cloud Platform (GCP) for scalability and accessibility.- Developed a user-friendly Streamlit front-end for intuitive interaction and data exploration.

Pune, Maharashtra, In

Developed ensemble-based models for plant seedling classification and weed detection:- High accuracy: Achieved 95.32% recall and 99.7% overall accuracy, outperforming existing techniques.- Novel approach: Combined shallow CNN for feature extraction with K-Nearest Neighbors for classification, refined through cross-validation.- Unique data augmentation: Implemented a technique using bilateral blurring, LAB colorspace conversion, segmentation, and image masking for improved performance.- Presented at conference: Work showcased at ICACIE'21, demonstrating its significance in the field.

Gurgaon, Haryana, In

Developed and launched a data-driven educational platform for 5,000+ users:Technology stack: Electron.js, React.js (front-end), Node.js(backend)Impact:- Increased user engagement by 40% through personalized learning paths and interactive data dashboards (Tableau).- Enhanced personalization: Utilized machine learning models (recommendation systems) for tailored learning paths.- Improved user experience: Conducted A/B testing and optimized UI elements (ReactJS, CSS, Bootstrap).Key achievements:- Designed engaging data visualizations to track user progress and platform usage.- Integrated machine learning for personalized learning recommendations.- Optimized user experience through A/B testing and UI/UX improvements.- Increased user engagement and platform usage with data-driven insights.

Master'S Degree, Computer Science

Bachelor Of Engineering - Be, Computer Science Engineering

Higher Secondary, Pcm