Inspiration

Vera is named after my late great-grandmother and she was built for my grandmother.

My grandmother still lives at home. Like a lot of older adults, her days can get quiet. Less activity leads to less energy, which leads to even less activity. My Granny is one of the most special people in my life and I want to do everything I can to help her live with vitality for as long as possible. That turned into a simple goal: build something that keeps her mind and body busy without feeling like a chore or a medical app.

Vera started as CALL-E based, as my entry in an earlier hackathon challenge: a scheduled phone call service where an AI companion called older adults to check in, chat about their day and talk them through a short daily exercise. That CALL-E project won a prize, but the most important test happened at home.

From phone calls to progress tracker

The demo was good enough to be recognized, but it wasn't ready for real life, and it showed me where I needed to step up. So I started the daily check-ins myself, by phone and text, with my real grandmother and turned the idea behind Vera into a progress tracker for her daily activity.

She couldn't pull up the tracker on her own, so every day she told me what activity she'd done. I added it to her log and then I sent her a screenshot. Within about a month, she went from 5 to 10 minutes on her stationary bike for a couple days a week to as much as 33 minutes, almost every day.

Then came the real proof. We took a family trip out of state that she would not have been up for before: a two hour drive to the nearest airport, getting through the airport, hours in a cramped plane seat and getting around an unfamiliar town to keep up with all of us. Because she'd built up her energy and stamina, we all had a wonderful time. Now she can't imagine missing another family trip.

The check-ins helped but what made the difference was simpler: seeing her progress and knowing her family could see it too.

That changed how I thought about Vera. An AI phone call happens once and is gone. It's easy to ignore and that's if it doesn't get caught by a spam filter first. A companion that lives in the home, remembers you, celebrates your streaks and brings your family's voices into the moment you reach a goal could do much more.

So for this hackathon, I rebuilt Vera for Alexa+.

What it does

Vera is a warm purple elephant companion for older adults, built for Alexa+ and designed after the Echo Show.

  • Games: word recall, matching pairs, finish the saying, and word categories, playable by tapping or just by talking, with more games on the way. In Chat, Vera can also play classic voice games like 20 Questions and riddles.
  • Exercise: guided chair routines and stretches, plus logging for anything active, from a shopping trip with lots of walking to gardening. Log by voice, by tapping, or with a guided walk timer.
  • Meditations: guided sessions that only count when they're finished. Right now there's one "Healing Affirmations" meditation, with more planned from 2 to 20 minutes.
  • Progress: a five week calendar of active days, streaks with a forgiving grace day, and badges. Missed days are never shamed.
  • Cheers: family members get one private link by text and are encouraged to record at least five 15 second cheers: voice, voice with a photo, or video, in their own words or from Vera's suggestions. Some cheers unlock when she reaches a goal and others play during an activity to keep her going. Vera plays at most two a day, so if five family members send five cheers each:

25cheers2perday12daysofencouragement

  • Chat: open conversation with optional starter chips like "Let's talk about the old days," and memory so Vera can ask about the things that matter to her.
  • Living room display: "Put my workout on the TV" hands the session to a big screen (a simulated Fire TV display).

How I built it

Every feature is a tool on a self-hosted MCP server (spec 2025-11-25, Streamable HTTP), the open standard behind Alexa+ integrations. The web app simulates the Alexa+ experience on an Echo Show style screen, and both the app and Amazon Nova call the same MCP tools.

  • Amazon Bedrock (Nova Pro): conversation and tool calling through the Converse API
  • Amazon Polly (Ruth neural voice): Vera's voice, slowed slightly for older listeners, with mouth movement synced to the audio
  • Amazon S3: exercise videos, meditations, and family cheer recordings
  • Amazon SES: "almost there" and "running low" emails to family
  • AWS End User Messaging SMS: automatic texts, built and ready behind a switch until a toll-free number is registered
  • Amazon Lightsail: hosting at vera.sha-i.com with HTTPS, an access code for real families, and a no-code demo for judges

I built it with Claude Code, in phases, with 80+ automated tests.

What changed from CALL-E

CALL-E was the starting idea. Everything in this repo was built during the submission window.

CALL-E Vera for Alexa+
Scheduled phone calls A companion that lives on the screen at home
One call type at a time Games, exercise, meditation, chat, and progress in one place
Reports sent to family Family takes part with cheers that unlock at real milestones
No memory between calls Remembers details and follows up on plans
Voice only Voice first, with large touch screens and a character

Challenges

  • Intent vs. action. When Vera suggested a walk and I said "I will walk today," the model logged a walk, finished the weekly goal, and played a family cheer. I fixed it in three layers: clearer instructions, a server check that rejects logging without a clear past tense completion and a new planning tool, so Vera says "Let me know when you're back" and follows up later.
  • Conversation quality. Nova Lite was fast but lost the thread in open conversation. Moving chat to Nova Pro, plus rules like "always respond to what she just said first," made a big difference.
  • Listening. Speech recognition cut people off at the first pause, which is exactly how older adults talk. I rebuilt it to keep listening through pauses, with a big "I'm done talking" button. I also removed auto-listening, so she taps the mic before speaking and Vera doesn't react to background sounds.
  • Honesty in a companion. Vera once invented a story about my real family and told it as true. Now she never makes up memories about real people.
  • Cloud setup. A Claude model wasn't available on my new AWS account, every service needed its own IAM permission, and App Runner turned out to be closed to new customers, so I moved hosting to Lightsail. All of it is in my friction log.

What I learned

  • Building for older adults means designing for dignity: big text, no shaming, and never talking down.
  • The hardest part of a voice companion isn't the AI, it's listening well: knowing when someone is done talking, and telling a plan apart from something already done.
  • MCP made Vera portable. Because every feature is a tool, the same Vera works on a web screen today and can plug into Alexa+ as it opens to more developers.
  • Most of all: progress means more when someone you love is cheering you on.

What's next

  • Connecting to Alexa+ on real Echo devices, plus a classic Alexa skill so my grandmother can use Vera on her own Echo now
  • A native Fire TV app for workouts and games on the big screen
  • More games and meditations
  • Automatic family texts once a toll-free number is registered
  • Gentle, consent based activity check-ins with Ring
  • Voice based cognitive screening, developed with clinicians

Built With

Share this project:

Updates

Submission history