AI Care

• 

  Logo

Inspiration

We built AI Care around a problem that feels deeply familiar in Singapore: the invisible “second shift” of caregiving. Many working adults are supporting ageing parents while also juggling full-time jobs, kids, and their own households. In most families, that coordination work quietly falls onto one person.

We were inspired by the idea that eldercare is not just a medical problem. It is also a communication, logistics, and fairness problem. Instead of building yet another tool that only monitors the elderly parent, we wanted to build something that helps the whole family share the load more transparently and sustainably.

What it does

AI Care is a Telegram-based caregiving assistant for families with ageing parents.

It uses a warm AI voice persona, Aunty May, to send medication reminders and symptom check-ins to the parent in Mandarin or English through voice messages. The parent can reply naturally by voice, and the system transcribes and logs those responses.

If medication is missed or left unconfirmed, AI Care escalates the task to the family group chat and routes it to the on-duty caregiver for that day. It also keeps a record of nudges, responses, and follow-through so the invisible caregiver workload becomes visible. Before a clinic visit, it generates a concise GP-ready briefing summarizing medication adherence, recurring symptoms, and key recent events.

How we built it

We built AI Care as an agentic workflow on top of tools families already use, instead of asking them to learn a brand new app.

The project is built in Python with FastAPI for the webhook server and python-telegram-bot for the messaging layer. We used APScheduler to run timed medication reminders, follow-ups, and recurring family updates.

For voice features, we used ElevenLabs for both text-to-speech and speech-to-text, which lets elderly users interact through voice instead of text-heavy interfaces. For intelligence and summarization, we used the OpenAI API to help compile structured GP briefings from recent family and symptom data.

We used Supabase/Postgres to store family setup, medication schedules, appointments, event logs, and cached audio assets. We also added PDF and QR generation so the family can bring a clinic-ready summary to short GP visits.

Challenges we ran into

One of the hardest challenges was designing for trust and safety. We had to make sure the AI feels warm and helpful without being deceptive or pretending to be a medical professional. That meant building clear disclosure, careful escalation rules, and strict guardrails around medical advice.

Another challenge was building a workflow that works for older users. Voice is much more natural than forms, but it also creates harder technical problems: multilingual replies, partial confirmations, noisy audio, and making sure the system understands what was said without over-claiming certainty.

We also had to think carefully about privacy and dignity. The parent’s actual voice messages stay private, while the family group only sees what they need operationally: whether something was confirmed, missed, escalated, or resolved.

Accomplishments that we're proud of

We are proud that AI Care is not just a chatbot concept, but a working caregiving workflow built around a realistic family dynamic.

We are especially proud of:

• Building a voice-first interaction model that feels accessible for elderly parents
• Turning missed medication reminders into coordinated family action instead of silent failure
• Making hidden caregiver labor visible through shared family tracking
• Generating a compact GP briefing that turns fragmented daily care data into something useful during a short clinic visit
• Creating a product that is emotionally aware and culturally grounded without crossing into manipulative “companion AI” territory

What we learned

We learned that eldercare technology should not only focus on the elderly user. In many cases, the real product challenge is reducing the coordination burden on the family around them.

We also learned that AI is most helpful here when it plays a narrow, respectful role: reminding, summarizing, routing, and escalating, rather than trying to act like a doctor or replace family judgment. Good AI in caregiving is less about sounding impressive and more about being dependable, legible, and safe.

What's next for AI Care

Our next step is to make AI Care more complete as a family caregiving platform.

We want to expand language support beyond Mandarin and English, improve appointment management, and make briefing generation even more useful for polyclinic workflows. We also want to build stronger caregiver analytics so families can better understand how responsibilities are being shared over time.

Longer term, we see AI Care becoming a practical coordination layer for ageing-in-place: helping families manage medication, symptoms, appointments, and caregiver load in one system while keeping humans firmly in control.

Built With

• apscheduler
• elevenlabs
• fastapi
• icalendar
• openai-api
• postgresql
• pydantic
• python
• python-telegram-bot
• qrcode
• reportlab
• sqlalchemy
• supabase