Inspiration

Healthcare facility data is rarely clean enough to use directly. Duplicate records, missing contact info, inconsistent locations, and partial evidence make it hard to know which facilities need human attention first.

We wanted to turn that problem into a workflow that felt operational instead of purely analytical: something a reviewer could actually use to decide what should be checked now, what can wait, and what was changed by a human.

What it does

Healthcare Readiness Desk turns 10,088 facility rows into an auditable triage queue. It scores each record for evidence, contact, geospatial, and operations coverage, hides likely-clean rows by default, surfaces suspicious facilities first, and lets reviewers:

inspect raw evidence
save review decisions to a separate state table
canonicalize duplicate families into one master record
export or hand off queues for operations
see before/after quality impact in the dashboard

How we built it

We built the app on Databricks Apps with an AppKit React frontend and Databricks SQL-backed queries. The workflow is:

Read the facility tables from Databricks.
Score each row for readiness and review urgency.
Auto-skip likely-clean rows.
Surface high-priority and duplicate-prone facilities.
Save reviewer decisions into a workspace review-state table.
Show state, district, and before/after impact views so the cleanup story is visible.

The app also includes bulk review actions, CSV export, clipboard handoff, and seeded example reviews so the demo feels alive instead of empty.

Challenges we ran into

The biggest challenges were:

handling messy source values safely without breaking the UI
making charts and summaries readable instead of overloaded
keeping source evidence read-only while still supporting write-back for reviewer decisions
dealing with Databricks permissions for the app, source tables, and review-state schema

We also had to harden the frontend so unexpected values would not crash the app. That mattered because real-world data is messy in ways that are not obvious from a sample.

Accomplishments that we're proud of

We’re proud that the app does more than display data. It actually supports a review workflow.

Highlights:

a ranked queue that prioritizes messy facilities first
duplicate canonicalization with a single action
persisted reviewer notes and overrides
a before/after impact report that makes the cleanup story visible
a live Databricks deployment that other workspace users can access

What we learned

We learned that the best data app is not the one that shows the most rows. It is the one that reduces cognitive load:

score first
triage second
keep raw evidence visible
store human decisions separately from the source data
show impact after the review loop runs

We also learned how important permission modeling is in Databricks. A working app is not just a UI problem; it is also a catalog, schema, and table access problem.

What’s next for Healthcare Readiness Desk

Next, we’d like to make the triage layer even smarter:

better canonicalization rules for duplicate families
threshold tuning based on reviewer behavior
reviewer assignment and queue ownership
stronger quality dashboards by state and district
export into downstream operations tools
automated checks for future incoming facility data

Built With

appkit
databricks-virtue-foundation-dataset-dais-2026.virtue-foundation-dataset.facilities-enrichment:-nfhs-district-indicators
india-post-pincode-directory-libraries:-@databricks/appkit
languages:-typescript
lucide-react
sql-frontend:-react
unity-catalog-storage:-workspace.healthcare-app.facility-review-state-data-sources:-databricks-clean.default.facilities-clean
vite-platform:-databricks-apps-data:-databricks-sql-warehouse