Helios

Inspiration

A heat warning can tell someone that conditions are worsening, but it does not establish whether they can afford cooling, reach a suitable place, or get help.

Helios began with this practical question: “What can a resident realistically do during a hot period?” Our initial focus is older adults with limited cooling access or mobility, alongside their chosen caregivers. We want to connect Singapore’s public environmental data to understandable choices that reflect a person’s circumstances.

What Helios will do

Helios is our proposed response to B1 — HeatGuard: Urban heat risk mapping for vulnerable communities.

The experience will follow three steps: prepare, choose an action, and ask for help.

Residents will see current outdoor heat conditions and a proposed one- to three-hour outlook, with timestamps and uncertainty clearly displayed. An accessible mobile interface will use large text and plain-language explanations.

Residents can optionally indicate practical constraints, such as unavailable cooling or difficulty travelling. Helios will use reviewed rules to suggest relevant options and explain why they appear. Candidate cooling destinations will show opening hours, access conditions and verification status.

If an option is not feasible, the resident can request help from a chosen caregiver. A secondary, simulated Active Ageing Centre view will demonstrate how consented requests, acknowledgements and unresolved needs could be tracked.

A supporting map will connect weather observations, planning-area demographic context and destination locations, while making the limits of geographic coverage visible.

How we plan to build it

The prototype will use Databricks Free Edition. Scheduled Python notebooks will ingest NEA weather feeds and public population and geographic data into Delta Lake. Bronze tables will preserve raw records; Silver tables will clean and align them; Gold tables will contain forecasts, area summaries and destination information.

Unity Catalog will support governance and lineage. MLflow will track comparisons between simple forecasting baselines and a small gradient-boosted model. Databricks SQL, an analytical dashboard and an authenticated Databricks App will support the demonstration.

Our two-week scope is deliberately bounded: three environmental feeds, one pilot planning area, five candidate destination records and one complete resident-to-caregiver journey.

What we have learned so far

Our research clarified that outdoor heat, indoor conditions and an individual’s health risk are different questions. Public weather observations can support environmental forecasting, while aggregate demographic data provides community context.

We also learned that a listed facility is not automatically an available cooling option. Opening hours, eligibility, accessibility and the resident’s ability to reach it are essential to making a suggestion useful.

These findings shaped our emphasis on practical choices, transparent evidence and resident-controlled support.

Challenges we expect to address

The main challenges are obtaining sufficient historical WBGT data, aligning weather feeds with different stations and timestamps, reconciling geographic boundaries, and checking destination suitability.

We will test API connectivity and viewer access early in the Databricks workspace. Missing or stale data will remain visibly unknown, and unverified destinations will retain their candidate status.

How we will evaluate the idea

We plan to compare one- and three-hour WBGT forecasts against persistence and time-of-day baselines using later, held-out observations. Evaluation will include forecast error, missed high-heat episodes and false alerts.

Scenario-based usability testing will compare Helios with an advisory-only experience, measuring correct action selection and time to identify a feasible option.

Current stage and intended impact

This is a Round 1 concept proposal. Implementation, model evaluation and user testing are planned; no forecasting accuracy or health-outcome improvement is claimed yet.

The hackathon demonstration will use public data and fictional resident and caregiver records. Community outreach will be simulated. Helios will not claim to measure conditions inside homes, predict individual illness, or provide an operational emergency-response service.

Our intended impact is to help residents understand approaching heat, identify a workable response and recognise when they need support.