Projects

The working systems behind the writing.

These are not portfolio case studies. They are ongoing knowledge systems: tools, habits, and experiments that turn reading, clinical learning, and AI work into durable output.

GitHub projects

Public work that shows the direction.

A selected map of repositories that connect the writing here to actual code: medical AI systems, local-first health tools, clinical data workflows, and research experiments.

View GitHub profile
RAG prototypemedical_assistant

AI Medical Knowledge Base Assistant

Evidence-based medical Q&A system that combines local medical textbooks, ChromaDB retrieval, PubMed search, and MedGemma answer generation.

PythonStreamlitChromaDBOllamaMedGemma
  • Parallel retrieval from knowledge base, PubMed, and model knowledge
  • Chinese medical terminology optimization
iOS product prototypeCcare-app

ChronicCare

Privacy-first chronic medication management app for reminders, adherence tracking, health measurements, refill alerts, and local backup.

SwiftUISwift ChartsHealthKitLocal storage
  • Adaptive medication reminders and snooze flow
  • Local-first design with no required account
iOS prototypesleep-app

Sleep Assistant

A lightweight bedtime companion focused on one nightly path: tonight plan, breathing, soundscape, phone down, and morning review.

SwiftUIAVAudioEngineUserDefaults
  • Low-friction bedtime flow instead of a complex dashboard
  • Offline, local-data product direction
Data toolingbio-clean-agent

Bio Clean Agent

TypeScript clinical data cleaning system with quality scoring, duplicate handling, missing-value strategies, audit trails, API, dashboard, and CLI.

TypeScriptNestJSNext.jsVercel AI SDKTurborepo
  • ISO 8000-style data quality assessment
  • LLM-powered interactive cleaning workflow
Research codemedical-automl

Medical AutoML Research

Clinical cardiovascular prediction experiments comparing tabular-to-text Transformers, LLM in-context learning, and traditional ML baselines.

PythonuvOllamascikit-learnPyTorch
  • External validation and calibration analysis
  • Publication-oriented figures, results, and statistical tests
Clinical workflow prototypemed-decoder

CardioVoice

Real-time cardiology outpatient documentation system connecting ESP32-S3 audio capture, medical speech recognition, MedGemma reports, and physician review.

PythonFastAPIStreamlitPostgreSQLESP32-S3
  • Local-first clinical documentation pipeline
  • Hardware plus AI workflow experiment

Project map

How projects connect to tracks