Research Assistant in Data Science & Educational AI at UNC School of Education, building open-source AI tutoring infrastructure for mathematics education.
About
Background
MS in Data Science Engineering (GPA 3.7) from the University of Houston, with a B.Tech in AI & Machine Learning.
I build across the full AI stack — from training deep learning models and designing multi-agent CrewAI pipelines to deploying production Streamlit apps backed by vector databases.
Currently a Research Assistant in Data Science & Educational AI at UNC School of Education, building reusable, open-source AI tutoring infrastructure for mathematics education.
Beyond AI, I ship full-stack systems end-to-end — a Next.js/TypeScript/GraphQL/Kubernetes app (FlowBoard) and a React/Node/PostgreSQL platform (DevLink) — to stay sharp across the whole stack, not just the model layer.
3.7
MS GPA
5+
Projects Shipped
88%
CNN Accuracy
40%
Grading Effort Saved
Experience
Where I've Worked
Aug 2026 – Present
Research Assistant, Data Science & Educational AI
UNC School of Education — Chapel Hill, NC
- Supporting development of reusable, open-source AI tutoring infrastructure for mathematics education.
- Working across data preparation, analysis, model evaluation, software development, and documentation.
- Collaborating with an interdisciplinary team spanning data science, AI, learning sciences, and education.
- Extending prior PhET Learning Progression feedback pipeline work into a broader open-source tutoring platform.
Aug 2025 – May 2026
Research Assistant
University of Houston — Houston, TX
- Built an AI chatbot integrated with PhET physics simulations to analyze 100+ student interaction responses.
- Designed Python pipelines to classify responses via Learning Progression rubrics — improving accuracy by 20%.
- Implemented LLM evaluation pipelines with self-consistency voting, boosting reliability by 25%.
- Automated personalized feedback generation, cutting manual grading effort by 40%.
- Built scalable processing workflows, improving pipeline efficiency by 30%.
May 2025 – Aug 2025
Machine Learning Intern
Ecare Medical Group — Houston, TX
- Built ML models using Scikit-learn and Pandas to analyze 5,000+ anonymized healthcare records.
- Performed feature engineering on 50+ clinical features, improving model performance by 15%.
- Trained Random Forest and Gradient Boosting models achieving 85%+ prediction accuracy.
- Created data visualization dashboards, improving reporting efficiency by 30%.
- Optimized preprocessing pipelines, reducing data processing time by 25%.
Education
Academic Background
MS Data Science Engineering
University of Houston, Texas, USA
B.Tech — AI & Machine Learning
Swarnandhra College of Engineering and Technology
Technical Skills
What I Work With
AI / LLM / GenAI
ML & Deep Learning
Languages & Backend
Data & Databases
Frontend & UI
Tools & Platforms
Featured Work
Projects
AI Medical Health Assistant
Full-stack AI assistant extracting 20+ medical values from PDFs/images, generating personalized insights with ~90% accuracy.
- 3 CrewAI agents cut report interpretation time by 70%
- LangChain + ChromaDB for 50+ patient records & trend analysis
- Streamlit UI with REST APIs for real-time insights
CrewAILangChainGPT-4o
ChromaDBStreamlitPyMuPDF
AI PhET Simulation Chatbot
AI chatbot integrated with PhET physics simulations to evaluate student responses and automate grading feedback at scale.
- LLM evaluation pipelines improved reliability by 25%
- Automated feedback reduced evaluation time by 40%
- Scalable Python modules for large-batch processing
LangChainOpenAI
PythonLLM EvalFastAPI
Diabetic Retinopathy Detection
CNN model classifying retinal images into severity stages with 88% accuracy — enabling automated early detection.
- Preprocessed and augmented 1,000+ retinal images
- Improved generalization by 15% via hyperparameter tuning
- 88% accuracy across all retinopathy severity stages
TensorFlowCNN
Computer VisionPythonNumPy
FlowBoard
Kanban-style project board built to demonstrate production TypeScript, GraphQL, and container orchestration end-to-end.
- Apollo GraphQL API layer over a Next.js/TypeScript frontend
- Containerized and deployed via Kubernetes
- CI/CD pipeline automated with GitHub Actions
Next.jsTypeScriptApollo GraphQL
KubernetesGitHub Actions
Full-stack developer networking platform built from scratch to close resume gaps in system design and scalable backend architecture.
- React frontend on Vercel, Node/Express backend on Render
- PostgreSQL (Neon) + Redis (Upstash) for persistence and caching
- GPT-4o-mini API integration for smart features
ReactNode.jsExpress
PostgreSQLRedisGPT-4o-mini
Research
Research & Thesis
Learning Progression-Guided AI Tutoring Infrastructure for Mathematical Sensemaking
MS thesis building an AI-assisted learning progression (LP) feedback system on PhET Coulomb's Law simulations. A two-LLM pipeline (accuracy gate + feedback generation agent) classifies 100+ student responses against LP rubrics and generates personalized feedback, with Supabase-backed persistence and self-consistency voting for reliability.
Security Approaches for Advanced Traffic Management Systems (ATMS)
Research on data analysis for security and privacy in intelligent traffic systems — identifying vulnerabilities in Intersection Signal Attacks (ISA) and designing protective mechanisms against exploitation vectors in smart city infrastructure.