National Breathe Free
Software Engineer
June 2025 — Present
National Breathe Free runs multi-facility healthcare operations software — analytics, collaborative reporting, and clinical data movement from fragmented office systems into cloud pipelines. My work spans the TypeScript product platform and the Python/AWS side of extraction and transform, with an emphasis on migrations, correctness, and systems you can debug when they break
- TypeScript
- Next.js
- PostgreSQL
- Drizzle
- Supabase
- Yjs
- Python
- ODBC / MSSQL
- S3
- SQS
- Protobuf
- Vitest
- DataDog
01
Platform migrations & architecture
Large production changes need blast-radius awareness, not just feature delivery. I owned a Drizzle ORM v1 migration that touched relation APIs, query shapes, migration tooling, and environment safeguards — then tightened the architecture so tests and layer boundaries could keep pace with the codebase
- Migrated a production TypeScript/PostgreSQL app to Drizzle ORM v1 across a 102-file blast radius, clearing 521 TypeScript errors while modernizing relation/query APIs and migration tooling
- Reworked testing from DB-coupled action-level suites into isolated use-case unit tests, enabling Vitest in CI with parallel lint, type-check, and test jobs
- Developed custom ESLint rules enforcing clean-architecture boundaries (server actions → use cases → persistence) and blocking direct database access outside the data layer
02
Clinical data pipelines
Clinic data lives in fragmented AdvancedMD Databases. I built and extended the path that pulls AdvancedMD/MSSQL clinical streams over ODBC, lands partitioned Apache Parquet in S3, and feeds downstream transform/control-plane work — with checkpointing so long runs can restart safely
- Built a Python clinical extraction service with checkpointed incremental ODBC pulls and partitioned Parquet landing in Amazon S3 for downstream transformation
- Implemented transform workers (shared ownership) that consume S3/SQS claim-check messages, transform Parquet artifacts, dual-write to Supabase targets, and honor pause/resume/shutdown control-plane commands
- Built Next.js control-surface features for DynamoDB-backed pipeline configuration and SQS/protobuf command operations so extractor/transformer fleets can be inspected and controlled safely
- Defined Buf-managed Protocol Buffer contracts for the pipeline control surface, generating Python and TypeScript clients with linting, breaking-change checks, and CI drift detection
03
Realtime reporting & operational workflows
Operators need shared, attributable report edits — not siloed spreadsheets. I worked across PostgreSQL triggers, Supabase Realtime, and Yjs client state to keep collaborative editing trustworthy, and shipped audit and export paths that match how clinics actually work
- Stabilized realtime collaborative editing by fixing channel/topic mismatches and separating high-frequency cursor broadcasts from presence state across Supabase Realtime, PostgreSQL triggers, and Yjs
- Built a PostgreSQL trigger-based normalization pipeline that turns raw appointment feeds into patient, provider, referral, facility, and visit records with idempotent upserts and failure logging
- Implemented an end-to-end audit trail for operational report changes — actor, required reason, action type, and old/new values — through a layered TypeScript/Drizzle path that does not block the primary write on logging failure
- Built facility/date-filtered billing and operational report workflows with CSV, Excel, and multi-file ZIP export across domain, use-case, repository, and UI layers
04
Observability & delivery safeguards
Production healthcare workflows need signals when something drifts. I added frontend telemetry with a controlled intake path, and helped productionize notification delivery with idempotent fan-out where design ownership was shared
- Added DataDog RUM, user/error-boundary instrumentation, and a server-side telemetry intake proxy to a production Next.js application
- Implemented and productionized a role-scoped overdue-report notification pipeline using a transactional outbox, PostgreSQL cron, Resend, idempotent fan-out, and retryable delivery (shared design attribution)
- Added database-backed feature flags and application-wide flag context to support safer incremental rollout of admin and reporting workflows