ClinicalHours v2 — the AI-native volunteer management platform replacing legacy clinic software. Kanban inbox summarization, applicant screening, AMCAS activity writing.
1,000+ users · 9,500+ facilities · live Stripe revenue
Last updated · Apr 2026
Reproducible R pipelines (tidyverse, ggplot2) for transcriptomic + metabolomics analysis at the Suzuki Lab. Python automation of protein interaction assays at the Cho Lab.
NIH-funded · DOHaD · SARS-CoV-2 / influenza immune evasion
Last updated · Apr 2026
Macro markets and event-driven trading. The biology of aging. Model quantization and the soft ceiling of what fits in 16MB. Production agent design.
Reading list growing faster than time allows
Last updated · Apr 2026
An AI-native volunteer management platform replacing legacy clinic software.
Production AI healthcare workforce platform. 1,000+ users across 9,500+ facilities, with live Stripe revenue. Built solo on Next.js + Supabase/Postgres with row-level security; Claude-powered Kanban inbox summarization, applicant screening, and AMCAS activity writing.
An AI assistant for the visually impaired. Tap, and the world describes itself back.
Tap-to-capture pipeline using multimodal LLMs for real-time scene understanding and situational feedback. Voice synthesis delivers spoken descriptions in under 4 seconds across all devices. Live on the App Store.
Natural disaster intelligence. Predicts hazards from messy multi-source data.
Multi-source ingestion pipeline and visualization layer that interprets and organizes natural disaster signals to predict future hazards. Coordinated technical direction across the team to ship a functional intelligence system.
An LLM compressed into 16MB. Trained in ten minutes on 8×H100. For sport.
Parameter-constrained LLM under a 16MB artifact limit and a 10-minute 8×H100 compute budget. int6 quantization-aware training (STE) with sliding-window evaluation, custom learning-rate scheduling, and gradient accumulation.
I move between research labs and a startup. Both are training data.
Python·TypeScript·JavaScript·Java·SQL·C++·R
Next.js·React·Node·Express·Flask·FastAPI·tidyverse·PyTorch
Postgres·Supabase·Vercel·AWS·Docker·Git·Stripe·Anthropic API
I think about AI the way people once thought about the early internet — so I'd rather ship the wrong thing fast than wait for permission.
Open to interesting conversations and collaborations — research positions, AI engineering, and anything in healthcare or production ML. The fastest way to reach me is email.