This stack evolves as the product evolves. I care most about reliability, clarity, and shipping.
frontend (patient + doctor dashboards)
- TypeScript
- React + Next.js
- Tailwind CSS (fast, consistent UI)
- MD/MDX for notes-style content when useful
backend / APIs
- Next.js API routes (simple endpoints)
- Node.js services when needed
- Python (for data science + model experimentation)
- FastAPI (for ML/analytics services)
data layer (longitudinal + time-series)
- Postgres (core system of record)
- Time-series patterns (Timescale/Influx-style approaches depending on needs)
- Redis (caching + queues)
- Object storage for large files (labs, waveforms, documents)
streaming (real-time wearable ingestion)
- MQTT / WebSockets (device-to-cloud patterns)
- Kafka-style streaming when scale requires it
ML / analytics
- NumPy / Pandas
- scikit-learn (baselines + feature work)
- PyTorch (deep learning)
- Time-series modeling + signal processing workflows
- Evaluation-first mindset: baselines, ablations, monitoring
retrieval / knowledge (clinical + research)
- Vector search for research/clinical notes (pgvector/Chroma-style approaches)
- Structured knowledge layers (ontologies + KG patterns)
interoperability (clinical integration)
- HL7 FHIR concepts (resources, bundles, terminology)
- HAPI FHIR when implementing real endpoints
- OMOP concepts for observational datasets
observability (when it's "real-time health")
- Logging, metrics, traces
- Dashboards (Grafana-style monitoring)
- Alerting for pipeline failures and sensor drift
devops / deployment
- GitHub + Actions (CI/CD)
- Vercel / Azure / cloud hosting depending on the app
- Environment variable hygiene + secrets management
device / wearable (concept + integration direction)
- BLE and mobile-to-cloud relay patterns
- Firmware + sensor pipelines (as specs mature)
- Secure device identity + data integrity
my defaults
- Build something simple that works.
- Measure it.
- Improve the parts that matter most.
- Keep the clinician/user workflow sacred.