Inspiration
Identifying where the real, highest risk gaps in healthcare are across India — finding underserved districts that lack adequate healthcare infrastructure despite having high health risk scores.
What it does
Risk IQ combines ML-predicted district health risk scores with geospatial analysis to map healthcare facility accessibility across 463 medium/high-risk districts in India. It identifies 25 critical districts with no facilities within 30km and provides facility contact information for intervention planning through an interactive app and dashboard.
How we built it
Using Databricks lakehouse architecture with Unity Catalog for governance, ensemble ML models (GBM + Linear Regression) for risk prediction, Haversine distance calculations for facility mapping, Lakebase Postgres for sub-10ms app queries, and Delta Lake for reliable storage. Built with three notebooks orchestrated via Databricks Jobs with table update triggers.
Challenges we ran into
Free Databricks edition pipeline limitations requiring workarounds Yahoo email domain blocking causing authentication issues Databricks Apps unavailable initially, then port configuration issues CREATE SYNC TABLE to Lakebase failed — required manual CLI sync for all three tables ML challenges: feature engineering complexity, missing data handling, inconsistent district name spellings requiring fuzzy matching, optimizing 1.45M distance calculations Accomplishments that we're proud of Analyzing 463 districts across 27 states, mapping 1.45M facility-district records, identifying 25 critical gap districts, and building a complete data lakehouse architecture from marketplace data to production app — all by three complete strangers collaborating remotely.
What we learned
Always verify feature availability in free/trial editions before architecture decisions, maintain CLI-based backup plans for UI-dependent workflows, budget extra time for authentication debugging, and test pipeline components incrementally rather than end-to-end.
What's next for Risk IQ
Automate manual steps (Lakebase sync, authentication workflows), expand to medium/low risk districts, integrate real-time facility status updates, build predictive models for facility demand, and add demographic overlays (population density, income levels).
Log in or sign up for Devpost to join the conversation.