Amazon
Coming soon.
JUNE 2026 - AUG 2026
Mayo Clinic
'Histology-Based Deep Learning Stratifies Progression-Free Survival in Trastuzumab Deruxtecan–Treated HER2-Negative Breast Cancer' accepted to 2026 Mayo Clinic AI Research Summit and 2026 Pathology Visions Conference. Built large-scale data pipeline processing 9,000+ DICOM medical imaging slides; optimized MIL (Multiple Instance Learning) system architecture for breast cancer biomarker prediction deployed at clinical scale. Engineered scalable inference pipeline for UNI foundation model generating survival outcome predictions with attention-based interpretability.
OCT 2025 - PRESENT
Georgia Institute of Technology
“Counterfactual Explanations for Agentic Workflows” accepted to HCXAI @ CHI 2026 (Poster Spotlight). Designed interactive interface and LLM-based assistant to visualize execution traces, generate counterfactuals, and explain workflow behavior, improving interpretability of complex multi-step systems.
JUNE 2025
Georgia Tech Research Institute
Selected as the top 5% of 1200+. Designed and implemented an inversion-based adaptive noise cancellation system using signal generators, spectrum analyzers, and tunable band-pass filters. Developed a real-time digital signal processing (DSP) pipeline with a statistically significant 48% average SNR improvement. Integrated the system into a custom-built RF transceiver for performance validation under realistic conditions.
JUNE - JULY 2025
United Nations WAY
Selected as 1 of 1500+. Organized, judged, and moderated the largest UN hackathon to date with 1,700+ participants, $10,000 in prizes, and attendance from UN Ambassadors and Directors. Revamped the WAY website for enhanced UX, accessibility, and mobile responsiveness across 50+ pages. Launched a global youth policy map with scraped data and custom fact sheets from 190+ countries. Streamlined onboarding for 1,300+ volunteers and restructured the content architecture for 250+ resources.
JAN 2025 - PRESENT
Stanford Department of Anesthesiology
Secured $5,000 in research funding. Extracted structured and unstructured EHR data (labs, flowsheets, medications, ICD10, vitals) to derive frailty scores, comorbidities, medication classes, and physiological predictors using SQL, Python, and BigQuery. Visualized variable distributions, performed imputation strategies, and constructed Table 1 stratified by blood pressure categories with standardized mean differences. Deployed and statistically interpreted logistic regression across 4,100+ patients.
JAN 2025 - PRESENT
Oracle
Led full-stack Oracle APEX app development, from schema design to UI, for a production billing platform using Oracle Cloud Infrastructure (OCI), APEX, SQL, and PL/SQL. Implemented FIFO debt resolution logic, dynamic error handling, and delinquency prediction modules. Conducted independent research on multi-class ultrasound cancer detection using ResNet18 with attention, improving classification accuracy across tumor types.
AUG 2024 - MAY 2025
Stanford Center for Artificial Intelligence in Medicine and Imaging
Selected as 1 of 20 nationwide. Trained a ResNet-101 image classifier on weak labels extracted from radiology reports via rule-based NLP using SpaCy and RadGraph. Built a custom dataset with transformation pipelines to reduce label bias and eliminate overrepresentation. Achieved 89% classification accuracy on pulmonary ultrasound data and placed 2nd in the program challenge.
JUNE - JULY 2024
Microsoft
Prototyped an iOS and Flutter app and NFC wearable for HIPAA-protected medical data transfer in emergencies. Programmed core NFC logic in Swift, built cross-platform UI in Dart, and explored offline storage with SQL. Led Scrum-based project management, coordinated UI/UX, and pitched the project to Microsoft and AT&T executives.
JULY 2023 - APR 2024
TSI + Cornell University
Selected as Top 10% of ~1,000. Adapted Brain Network Transformer using Python GNNs for multi-state neuroimaging recognition. Published findings in the Talaria Journal and presented at the Talaria 2023 International Conference.
JUNE - AUG 2023
Midwestern University
Engineered POCUS-Net, a novel transformer-based model to automate gastric ultrasound analysis. Classified gastric contents (liquid, solid, air) and segmented diameters to estimate operable volume. Achieved ~87% accuracy and integrated volume-based decision weighting to output aspiration risk.
AUG 2023 - APR 2024