Inspiration
Singapore officially reaches "super-aged" status in 2026, with 1 in 5 residents aged 65 and above. Social isolation among these seniors is a silent health crisis. Currently, outreach by the Silver Generation Office and Active Ageing Centres (AACs) is highly manual. Grassroots coordinators cannot physically knock on a million doors, and they lack a data-driven way to prioritize which neighborhoods need immediate intervention. We estimate there are roughly 110,100 seniors living alone across 28 HDB-residential planning areas,these are the individuals who need proactive care the most.
What it does
SilverWatch is a batch risk-scoring platform that identifies the invisible micro-clusters of vulnerable seniors. It evaluates 28 HDB-residential planning areas using a custom Social Isolation Risk Index (SIRI). Instead of just plotting where seniors live, it outputs a live Databricks AI/BI Dashboard that ranks areas into Priority Tiers, giving social workers a Top-10 targeted outreach list to deploy their limited volunteer hours effectively.
How we built it
We built a Medallion architecture entirely on Databricks Free Edition.
Data Ingestion: We pulled three open datasets from data.gov.sg: HDB Elderly demographics, SingStat Living Arrangements, and MSF AAC capacities.
ETL & Storage: We used PySpark in Databricks Notebooks to ingest, clean, and join the data, storing it in Delta Lake (Bronze, Silver, Gold tables) governed by Unity Catalog.
Scoring Logic: We engineered the SIRI score using a transparent, weighted composite formula: 40% Living-alone ratio, 35% Senior volume, and 25% AAC coverage deficit. Areas scoring >= 70 were assigned to Tier 1: Critical Outreach.
Visualization: We queried the Gold tables using Databricks SQL to feed our final AI/BI Dashboard visualizations.
Challenges we ran into
Working strictly within the Databricks Free Edition serverless compute quotas meant we had to be highly strategic. Instead of building a compute-heavy real-time streaming pipeline (which risked locking us out of our workspace), we engineered a reliable, batch-only design. Furthermore, we had to choose between using a "black-box" machine learning model versus a weighted formula. We opted for the formula because government social workers require an explainable, auditable metric (line-by-line) to justify resource allocation.
Accomplishments that we're proud of
We successfully narrowed the operational search space by 79%, filtering 28 areas down to 6 Tier 1 Critical zones (led by Bukit Merah and Queenstown at ~85). Our prototype mathematically proved that our Tier 1 areas average 1.6x more seniors per AAC than our Tier 3 areas, validating that our index accurately identifies severe infrastructure coverage gaps.
What we learned
We learned how to orchestrate a full Delta Lake Medallion architecture, and more importantly, we learned that for public-sector AI tools, transparent data rigor and explainable math are often much more valuable to end-users than implementing complex machine learning just for the sake of it.
What's next for SilverWatch
If selected for the two-week mentored sprint, our goals are to run weight-sensitivity tests tracked in MLflow (to mathematically validate our current expert-weighted index), deploy a Databricks Genie space so non-technical volunteer coordinators can query the dashboard using natural language, and drill our macro planning-area data down to block-level micro-clusters.
Credits & Honesty Notes
Built on Databricks Free Edition using the DAISI template. Datasets sourced from data.gov.sg. We utilized Databricks Genie to assist in structuring the pipeline logic. Note: The 110,100 figure is estimated from living-alone ratios, and current scores are an expert-weighted index not yet validated against ground-truth isolation outcomes.
Team
Asher Heng (Temasek Polytechnic, Diploma in Applied Artificial Intelligence, Year 2)
Muhammad Adi Asyraf Bin Mohamed Saleh (Temasek Polytechnic, Diploma in Applied Artificial Intelligence, Year 2)
Taufik Ahmed Niloy (Republic Polytechnic, Applied AI and Analytics, Year 2)
Nuh'Aiman (Republic Polytechnic, Cyber Security and Digital Forensics, Year 2)
Built With
- ai/bi-dashboards
- databricks
- databricks-sql
- delta-lake
- genie
- pyspark
- python
- unity-catalog