Databricks Hackathon Hub

All information related to the Databricks Apps & Agents Hackathon for Good can be found here: developers.databricks.com/hackathon

Announcements and questions related to credits, the Databricks platform, or Virtue Foundation can be found in the hackathon's Discord during event hours: discord.com/invite/bedRGCjFq

Challenge

You are given 10,000 messy records of healthcare facilities across India. Each record includes structured fields such as location and specialties, plus uneven free-text descriptions of claimed capabilities, procedures, equipment, and services.

Build a Databricks App that helps a non-technical healthcare planner, NGO coordinator, or analyst turn this messy data into decisions they can trust.

Your app should extract useful structure from the records, show evidence for its conclusions, communicate uncertainty honestly, and let users save or revise their work.

Tracks

Track 1: Facility Trust Desk

Question: Can this facility actually do what it claims?

Build an app that evaluates facility claims for capabilities such as ICU, maternity, emergency, oncology, trauma, or NICU. For each facility and capability, produce a trust signal such as strong evidence, partial evidence, weak or suspicious evidence, or no claim.

Track 2: Medical Desert Planner

Question: Where are the highest-risk gaps in care, and how confident are we that those gaps are real?

Build an app that aggregates trust-weighted facility evidence across geography, such as state, city, district, or PIN code. Help planners distinguish real care gaps from data-poor regions.

Track 3: Referral Copilot

Question: Where should a patient or coordinator actually go?

Build an app where a user enters a location and a care need, such as “dialysis near Jaipur” or “emergency surgery near Patna,” and receives an evidence-attached shortlist of candidate facilities.

Track 4: Data Readiness Desk

Question: What needs to be fixed before this dataset can be trusted for planning?

Build an app for profiling, reviewing, and improving the facility dataset. Surface contradictions, suspicious claims, sparse fields, and high-leverage records for human review.

Dataset

The provided dataset contains 10,000 Indian healthcare facility records and 51 columns.

All records include facility name, state, city, latitude, longitude, controlled specialties, a description, and source URLs; 9,996 records include a postcode. The extracted evidence fields are noisy, repetitive, and unevenly supported:

Useful evidence appears across description, capability, procedure, equipment, specialties, and source_urls. Teams should treat these fields as claims to verify rather than ground truth.

• Product judgment
  Is the user clear? Are the workflow and tradeoffs thoughtful?
• Evidence and Uncertainty
  Are outputs grounded in citations? Is uncertainty handled honestly?
• Technical Execution
  Does the app work reliably in a live demo? Are Databricks capabilities used well?
• Ambition
  Did the team go beyond the minimum workflow in a meaningful way?