CareReach

In India, specialist doctors cluster in cities while the districts with the highest health burden — maternal anaemia, child malnutrition, undiagnosed hypertension and diabetes — often have no specialist at all. Visiting-specialist and outreach programs can reach those places, but they're scarce: send one to the wrong district and a rare resource is wasted. CareReach makes sure each visit lands where the unmet need is greatest, so more people get specialist care that would otherwise never reach them.

Finding those places is harder than it sounds. The data lives in four public datasets (NFHS-5 health indicators, Census population, a facility directory, geoBoundaries polygons) that share no common key — only district names, which disagree across India's 700+ districts. Joining on names drops half the country and makes join failures look like real supply gaps. So we:

Gave every facility and district one canonical code, placing facilities by geometry (point-in-polygon on real coordinates, never guessed from an address) so "no facility" genuinely means none. Scored each district's unmet need = 0.50·demand (NFHS health need + population) + 0.30·facility-gap + 0.20·medic-fit. Today the medic-fit term is a placeholder, so CareReach surfaces the specialist-care gaps — high demand, little or no ready care — with matching a specific medic as the next step. Shipped it as a Databricks App on AppKit (Lakeflow → Lakebase → React): an India map, a ranked worklist, and a fully transparent "Why this status?" drawer — so a health planner doesn't just see where to send help, but why. The payoff is simple: scarce specialist time, pointed at the districts where it saves the most lives

Built With

• claude
• codex
• databricks