AI for elderly people living alone

Background

I took part in a volunteering event where we cleaned elderly residents' homes. Seniors who had occasional help from a domestic helper were doing fine. The ones living alone were not. In the flat I was assigned to, something had stained the floor so badly that the cardboard laid over it had melted into the floor. There was no running water and no electricity. In other flats there were reports of faeces on the floor.

Conditions like that don't appear overnight. Along the way, people notice: a neighbour stops seeing someone at the lift, a utility account is cut off, a check-in call goes unanswered, a volunteer writes a worried note. But each signal sits with a different person or organisation, arrives in a different form, and is often uncertain on its own. Nobody sees the whole picture, so nobody acts until it's very late.

Who you are designing for

Elderly people living alone, first and always. Around them are the people who could help if the right information reached them in time: neighbours, volunteers, family members, community care and active ageing centres, town councils and social service agencies.

Many seniors are more comfortable in Mandarin, Malay, Tamil or a dialect than in English, and many won't use an app. Design for that.

What your system should do

1. Reconcile sources. Take in reports from different channels (a neighbour's message, volunteer visit notes, missed check-ins, a family member's call) and recognise when they're about the same person.
2. Weigh urgency and uncertainty. Separate "probably nothing" from "someone needs to knock on the door today", and show how confident the system is and why.
3. Recommend an outreach pathway. Who should respond, and how: a call, a home visit, or an emergency response.
4. Respect eligibility and capability. Only route to organisations that serve this person and can actually do what's needed.
5. Coordinate across organisations. When a case needs several things at once (a home visit, utilities restored, a deep clean), split the work and keep everyone in step.
6. Track to resolution. Follow every request from sent, to acknowledged, to resolved, and escalate when it stalls.

Ideas to explore

Optional. Strong teams pick one or two and go deep.

• Multilingual intake: voice notes or messages in any language, understood and summarised by Qwen.
• An agent that drafts the referral, sends it, and chases organisations that haven't acknowledged it.
• A coordinator view showing every open case, its urgency and who is holding it up.
• Explanations a human can check: which signals led to this urgency rating.

Things to think about

• Dignity and consent. These are people's homes and lives. What would you be comfortable with if it were your grandparent?
• False alarms vs missed cases. Which mistake is worse, and how does your design balance them?
• Humans in the loop. Where must a person confirm before anything happens?
• When the AI is wrong. What happens if it misreads a report or merges two different people?

What to deliver

A working prototype

Web or mobile, demoed live. The core journey has to actually work.

Real use of AI

AI that meaningfully contributes to the solution. Qwen through Alibaba Cloud Model Studio is supported, but any suitable model works.

Deployed and explainable

A live URL on Alibaba Cloud or any suitable platform, plus a simple architecture diagram you can explain.

A 10-minute presentation

Problem, demo, architecture, what's next. Every member ready for Q&A.

Use mock or synthetic data. Never put real personal information into your prototype.

How it's judged

Two judges score every team on this problem statement out of 100, using the technical track rubric.

1. Working prototype25 pts

   It runs live, end to end. The main user journey works without hand-waving.

2. AI capability & agentic workflow20 pts

   Does the AI meaningfully contribute to the solution? Where it fits, can it reason, use tools, retrieve information or take actions, rather than only answer questions?

3. System & cloud architecture20 pts

   A sensible technical architecture, and you can explain why each component or service is there. Deploy on Alibaba Cloud or any suitable platform.

4. Impact & problem fit20 pts

   Solves the brief for a real, specific user, rather than building technology for its own sake.

5. Pitch & demo15 pts

   Clear story within 10 minutes, a working demo, and team members who can answer technical questions.

+5 bonus if your prototype works in at least one Southeast Asian language besides English.

Judges reward appropriate use of AI, not complexity for its own sake. Using Alibaba Cloud, Qwen or Qoder is supported but not required to score well.

Draft · final weights are confirmed before the workshops. Compare both track rubrics