Clinical-Lab Job Decision Dashboard
Family-user-informed product design, frontend/backend, and operations
2026-07-01 — present
A mobile-first dashboard that normalizes job-platform and hospital-career-page sources, explains deadline/region/employment recommendations, and keeps account-scoped application state.
System Architecture
Problem Solving
Source-specific deadline, region, and employment formats make comparison and prioritization difficult
Normalize fields at ingestion and retain the collected source link, freshness, and open state in the read model
One screen provides deadline, region, employment filters, and recommendation reasons
Duplicate workers or transient failures can cause duplicate sends or missed postings
Separate a dedupe-key outbox, atomic lease, bounded retry, and watermark advancement after confirmed delivery
An explicit state contract that does not complete the posting stream before confirmed delivery
Synthetic scenarios
Deterministic fixtures reproduce the code paths and boundaries without operational or personal data.
Project Description
Adapters collect job platforms and hospital career pages into one normalized read model for posting ID, deadline, region, and employment type. Public discovery stays separate from account-scoped interest/application/interview/result state. Notifications handle duplicate workers and transient failures through dedupe keys, atomic leases, and watermark advancement only after confirmed delivery. The public case excludes real applications, family data, operating URLs, and notification destinations.
Highlights
- Multi-platform and hospital-source adapters with a normalized read model
- Explainable recommendations from collected source links, deadline, region, and employment type
- Account-scoped interest/application/interview/result tracker
- Durable outbox with dedupe, lease, and delivery confirmation
Tech Decisions
- ▶Normalize source-specific deadline text once at ingestion to keep query/filter semantics consistent
- ▶Separate public discovery from account-scoped application state
- ▶Tie outbox completion and watermark advancement to delivery confirmation to make the loss boundary explicit
Lessons Learned
- •A family-facing product earns trust by showing the collected source link and decision reason before automation.
- •Notification reliability starts with explicit completion and watermark boundaries, not retry counts alone.