👋 Hi, I’m @mansapatel111

• I'm Mansa, I'm a fourth year CS major at University of Florida, with interests in AI/ML and software engineering!
• I’m big on continuously learning and building my skills, and believe taking initiative is key to growth.
• Always looking to leverage my skills at hackathons, working on research projects or startup ideas - got any leads? Let me know! Linkedin

📫 How to reach me

• Email: mansapatel111@gmail.com

🎓 Education

• BSE in Computer Science, University of Florida, expected Spring 2026
• Minors: Mathematics and Engineering Innovation, AI/ML and Product Management Certificate

🌟 Key Skills and Achievements

• Languages: Python, JavaScript (React, Node.js), TypeScript, C++, C, Swift, HTML/CSS, ARM64 assembly

• Technologies: MongoDB, Neo4j, PyTorch, TensorFlow, Scikit-Learn, Postgre SQL, Pandas, NumPy, SciKit, Matplotlib, Tableau

• Tools: Notion, Figma, GitHub, Jira, Google Cloud Platform

• Best Zoom track at Stanford University's Tree Hacks (Selected as one of the 1000 hackers from 12,000+ applicants)

• Best AR/VR App at WingHacks

💻 Recent Projects

🎨 Art Beyond Sight - Multimodal AI Accessibility Platform: Repo

• Web App (Harn Museum of Art, Florida): https://youtu.be/sVN4-ctr4sM

• Mobile App (Carnegie Museum of Art, Pittsburgh): https://youtu.be/ilLqgHXgBVM?si=CrvhkGuisKiX2y3W

• User Demo (Jose’s Feedback): https://youtu.be/v9R-RLsbhaA

• What I did: Engineered a cross-platform mobile application transforming visual art into immersive auditory experiences through a sophisticated AI pipeline. Architected a three-stage workflow: (1) Mistral Navigator for vision analysis and contextual metadata extraction, (2) Suno AI for emotion-driven music generation mapped from visual elements, and (3) MongoDB caching layer for sub-500ms retrieval. Implemented WCAG AAA-compliant accessibility features including dual TTS systems (VoiceOver/TalkBack + Unreal Speech), adjustable speech parameters, haptic feedback, and screen reader integration with semantic focus management.

• Tech Stack: React Native, Expo, Overshoot, FastAPI, MongoDB Atlas, Mistral AI Navigator, Suno AI, Unreal Speech TTS, TypeScript, Python

🧠 Multimodal ADHD Prediction System - WiDS Datathon 2025: Repo

• What I did: Engineered a hybrid deep learning architecture for dual-task classification (ADHD diagnosis and sex prediction) using functional neuroimaging and behavioral data. Designed a Graph Neural Network to model brain region connectivity patterns from fMRI data, combined with dense neural networks processing socio-demographic features through embedding layers. Implemented selective imputation regression for systematic missing data handling and deployed XGBoost ensemble with Optuna-based hyperparameter optimization. Addressed severe class imbalance using SMOTE augmentation and achieved interpretable predictions through SHAP analysis, identifying executive control regions as key predictors. Built complete preprocessing pipeline ensuring feature alignment between training and test data.

• Tech Stack: PyTorch Geometric (GNN), XGBoost, Optuna, SMOTE, SHAP, scikit-learn, PCA, Stratified K-Fold CV, Python

🔬 Uncertainty-Aware Skin Cancer Classifier with Fairness Auditing: Repo

• What I did: Built a production-grade melanoma detection system combining deep learning with Bayesian uncertainty quantification and demographic fairness auditing. Implemented Monte Carlo Dropout (50 inference passes) to generate prediction confidence intervals and predictive entropy scores, enabling automatic flagging of uncertain cases for clinical deferral. Conducted comprehensive fairness analysis across sex, age groups, and anatomical locations to identify and mitigate demographic disparities in diagnostic accuracy.

• Tech Stack: PyTorch, EfficientNet-B0, Monte Carlo Dropout, Cosine Annealing Scheduler, Class-Weighted Loss, Gradio, scikit-learn

🌲 Noteworthy - Real-Time AI Lecture Visualization System: Repo | Devpost

• What I did: Developed a multimodal AI-powered lecture companion that transforms speech into real-time visual representations during live lectures. Engineered a full-stack pipeline integrating Zoom WebSocket technology for live audio capture, custom JSON-to-text parsers for dynamic transcript generation, and LLM-driven natural language processing to extract key concepts. Implemented OpenAI API integration to generate P5.js code for real-time sketch rendering, creating standardized visual representations of abstract concepts as lectures progress. Built an intelligent note-taking workspace with multi-source media integration (PDFs, chatbots), leveraging Perplexity AI for automated summarization and resource recommendations, and Gemini AI for context-aware Q&A. Designed collaborative features with Zoom Calendar API integration for automatic course organization and seamless note sharing across devices.

• Tech Stack: Zoom API (WebSocket, Calendar), OpenAI API, Perplexity AI, Gemini AI, P5.js, React, Node.js, Python, Playwright API, Ngrok, HTML/CSS, Bash, Postman

🏢 Technology News Insights Engine - Accenture AI Studio Challenge: Repo

• What we did: Developed an intelligent news analytics platform for Accenture's AI Studio Challenge through Breakthrough Tech AI to match startups with client needs. Engineered a graph-based sentiment analysis pipeline using Neo4j to model relationships between technology trends, companies, and news sentiment. Built a custom NLP processing system using SpaCy for named entity recognition and NLTK for sentiment classification, enabling automated extraction of key insights from technology news articles. Implemented machine learning models to track emerging tech trends and generate comprehensive visualizations identifying optimal startup-client pairings based on domain expertise, market sentiment, and innovation metrics.

• Tech Stack: Python, Neo4j (Graph Database), SpaCy (NER), NLTK (Sentiment Analysis), Machine Learning, Data Visualization

🛠️ Experience

• AI/ML Intern, DELL Technologies: Built an Agentic AI workflow automation system for Dell Life Cycle Hub service, streamlining asset data processing for 2k+ global clients. Developed a machine learning classification model to predict configuration matrices and categorize customer asset data. Designed a refined data categorization pipeline to speed up error processing by 30% for PSAS Troubleshooting AI Agent
• Teaching Assisstant, Software Engineering: Led discussion sessions for 35 + students and made project/lecture materials showacasing in depth software engineering fundamental knowledge.
• Machine Learning Intern, BreakThrough Tech AI: Worked on tech article analysis using Neo4j, sentiment analysis, and data visualization for actionable insights.
• Teaching Assisstant, UF CS Department: Led lab lectures, created exams for 700+ students, provided office hour support and personalized assistance to help students excel in computer science courses, with a focus on python programming fundamentals class.
• Research Assisstant, UF Health: Assissting in a research team focused on developing deep learning neural networks for cardiovascular disease prediction among patients.
• Fullstack SWE Intern, FLagIt: Developed frontend features for an emotional health tracking app called FlagIt, using Flask and React.
• Product Management Intern, Epic Hire: Developed figma designs for a recruiting platform designed for AVP and spatial computing platforms

🌏 Fun Facts

• Multilingual: Fluent in English, Hindi, Gujrati and Marathi.
• Interests: Passionate about travel and culture, have explored 10 countries across asia, and counting. I love learning about new cultures, trying different cuisines and learning about local history and heritage whenever I travel!