Research & Projects
NIH Network Analysis
Clustered 30,000+ NIH abstracts with BERTopic and visualized co-PI networks to reveal collaboration dynamics and funding inequality at UNC.
Spatial Transcriptomics
Benchmarked 3 graph-building frameworks (Seurat, Squidpy, Niches) on Stimage-1K4M, comparing model outputs and loss matrices for spatial gene expression analysis.
eNet Enhancer Network
Constructed enhancer network from single-cell multi-omics (scATAC + scRNA, 18k+ cells), identifying rare modular topologies (8.39%) predictive of MPA Leukemia driver genes.
Neuromatch Neuron Simulation
Decoded inter-regional brain communication via Leiden Analysis on the IBL dataset (30k+ neurons), predicting trial-by-trial behavioral decisions at 78% accuracy.
UNC Math Directed Reading Program
Working at the intersection of ODE/PDE, dynamical systems, and mathematical modeling. Watch this space!
CHIP741 Healthcare App
Final project: building a healthcare informatics app that makes clinical data more accessible and actionable. More coming soon!
Datathon & Hackathon
What I did
- Led a 5-person team to analyze 10k+ diabetic patient records in a 48-hour competition. Directed end-to-end analytical pipeline on a diabetes-anchored cohort, engineering longitudinal EHR features including inter-visit gap sequences, cv_gap irregularity metric, and ED utilisation flags with SDOH risk stratification across 8 social determinant domains.
- Surfaced structural health inequality via silhouette-optimised K-Means (k=3) and logistic regression — High-SDOH patients show 2.5× higher ED rates and concentrate 5× more in the ED-Dependent archetype (p<0.001). Counterfactual simulation quantified a 16.6pp risk reduction if structural barriers were removed.
- Translated findings into a geospatial mobile clinic intervention: optimised routing circuit covering 32,686 patients across 15 stops. Integrated team's Markov Chain comorbidity network and MLP risk-scoring dashboard into a unified live-pitch presentation.
What I built
- Built WageGuard, a bilingual AI wage-protection tool for Latino workers — combining voice-based case intake with automated wage violation detection in a single conversational flow.
- Developed React/Vite frontend and end-to-end agent workflow powered by Gemini 2.5 Flash + ElevenLabs TTS; designed the full conversational UX from scratch in 24 hours.
- Implemented real-time API integration & testing for reasoning and response generation, enabling live wage-law analysis with multilingual voice output.