• 

  Senior Dashboard

• 

  Family Dashboard

Inspiration

Many older adults live alone, and their families live far away. The worry is rarely about big emergencies. It's the small things: a missed pill, a skipped meal, a quieter voice than usual. Reminder apps depend on screens and rigid schedules, and a phone call every day isn't always possible.

Voice is the most natural interface for this group, and Alexa+ style assistants can now hold real conversations, remember context and take actions. We wanted to see whether that could become a gentle daily check-in that gives families peace of mind without being intrusive.

What it does

Check-in Buddy is a simulated Alexa+ experience in a web app. It:

• Holds a short daily voice conversation with an older adult about medication, meals, mood and activity
• Remembers previous days and refers to them naturally
• Takes actions through tools: logging medication, meals and mood, setting reminders, and notifying family
• Applies escalation rules, such as a missed medication two days in a row, so families get alerts only when they matter
• Shows a family dashboard with daily summaries, trends and alert history

It is a wellness check-in tool, not a medical device. It never diagnoses or gives medical advice, and it points users to a professional or family member when health questions come up. All demo data is synthetic.

How we built it

• Frontend: [HTML/JS or React] with the browser Web Speech API for voice input and spoken replies
• Backend: Python with FastAPI
• AI: [Amazon Bedrock ] with tool calling for medication logging, reminders and family notifications
• Memory and data: SQLite for daily logs and conversation history
• Escalation engine: rule-based checks that run over the logged data
• Evaluation: a suite of [50%] scripted conversation scenarios run through pytest, measuring correct tool use, escalation accuracy and safety behavior

Challenges we ran into

• Making speech recognition reliable enough for a smooth conversation, and what the text fallback looked like
• Getting the model to call tools consistently, and how structured outputs or retries fixed it
• Tuning escalation so it catches real concerns without false alarms
• Keeping the assistant from drifting into medical advice

Accomplishments that we're proud of

• A working end-to-end flow from voice check-in to family alert
• An evaluation suite of [10+] scenarios, with a measured [90%] pass rate on tool calls and [80%] on escalation accuracy
• A fully open-source repository that someone can run in under minutes
• Safety behavior that held up across adversarial test cases

What we learned Will be answered after the Tool comes in execution :

• Why evaluation matters as much as the demo for AI systems
• How memory changes a conversation from scripted to personal
• Designing for older adults: pacing, simple language and patience
• Where the line sits between helpful check-ins and overstepping into medical territory

What's next for Check-in Buddy

• Real Alexa+ integration if and when an appropriate path becomes available
• Multilingual support, starting with Hindi and Kannada
• Richer signals, such as changes in speech patterns or routine, with the user's consent
• Caregiver customization: choosing which alerts they receive and how
• A formal review of privacy, consent and safety with geriatric care and accessibility experts before any real-world use

Built With

• bedrock
• llm
• python
• react
• sqlite