June Lee
Research Engineer · Healthcare AI
I develop clinically aligned representation learning systems for biomedical AI. I move beyond black-box optimization by embedding physiological and structural priors into models to improve robustness under real-world distribution shift.
Research & Projects
Projects
WiseMind: Neurosurgical AI Platform
End-to-end clinical AI platform for neurosurgery, deployed at Harborview Medical Center. Integrates multimodal retrieval, spine analysis, and surgical simulation into a unified full-stack system.
TB Detection from Cough Sounds
AUC 0.9519, 92.7% sensitivity — acoustic TB screening reducing diagnostic latency from weeks to real-time. 57 physiological biomarkers from 7,031 cough segments.
Learn more ↗Vector Decomposition from Frozen Segmentation Encoders
Latent biomarker axes recovered from frozen encoders with zero trainable parameters. Centroid-difference projections achieve 5-fold cosθ = 0.90/0.96 directional stability. Zero-shot CVS AUC 0.803, PRL AUC 0.716. Validated via SWI gradient (p<0.001) and PRL rim (ρ=−0.49, p<0.001).
Learn more ↗Tissue-Aware Drug Response Prediction
R² = 0.947 (Pearson r = 0.977) with 3× MSE reduction over best baseline on 114,215 drug–cell line pairs (181 drugs, 682 cell lines, 13 tissues). Tissue weighting accounts for 89% of total R² gain (ΔR² = +0.148) via ROKU gene weighting from GTEx.
Google MedGemma 2026 Healthcare Hackathon
Three-agent multimodal clinical reasoning pipeline (Audio/HeAR, CXR/MedSigLIP, CT) with confidence-weighted late fusion — diagnosis in ~5s on consumer hardware. 3-level hierarchical reasoning with FHIR-compatible outputs for resource-limited deployment.
Academic Output
Publications
Let's Connect
Get in Touch
I'm always interested in discussing collaborations in Healthcare AI, medical imaging research, or any exciting biomedical engineering challenges.