9DSZKW - small usage limits big dreams - Amble

Inspiration

23% of Canadians aged 15 to 24 always or often feel lonely, the highest of any age group (Statistics Canada). Only 46% of Canadian adults get the recommended 150 active minutes a week (Statistics Canada). Making a friend usually means planning a coffee and making small talk, which is exactly the hard part for newcomers, for people living alone and for anyone shy. We wanted something that gets people outside and gives them something to do together besides small talk.

What it does

Amble answers Prompt 1 through two focus areas: Community Connection & Inclusion and Proactive Health & Wellness.

Tap "I'm free" and Amble pairs you with someone nearby who's free too. You both get the same small 20-minute quest at a real, public place in Calgary, like a game of foosball in the engineering building. There's something to do besides small talk, and a natural ending. Take a photo, and an AI referee checks it against the quest. Then you get a recap of your walk, and you choose whether to stay in touch. You only connect if you both say yes.

With friends, crews take on quests together, earn coins and race on a weekly board, and play Claim, a game where you take spots on the campus map by beating that spot's challenge.

The You tab shows numbers that matter, privately and never ranked: active minutes against Canada's 150-minute guideline, "familiar faces" (people you've done several quests with), new places, and how you felt before and after.

Inclusive and safe by design

Comfort settings: step-free places only, a short and easy pace, and "side by side" quests for people who'd rather not chat face to face.
You meet at a public quest spot, never at each other's location. Your partner sees only a rough distance, then a first name once you both say yes.
Reporting a problem is always one tap away, and a report means you're never paired again.
Your location is deleted the moment you're matched, and you can download or delete all your data from the privacy page.

Judges can try it alone at amble.health with "Try the demo": no partner, account or GPS needed. A practice partner, Priya, plays along.

How we built it

One Next.js (TypeScript) app on Vercel at amble.health. Auth0 handles sign-in. Postgres on Tiger Data stores everything, with location pings in a TimescaleDB hypertable and a 30-day retention policy. We wrote every quest by hand, at real public places around UCalgary and Calgary. Claude Haiku (Anthropic) checks proof photos (and videos, from three still frames) and writes the recaps. Every AI call has a timeout and a fallback, so nobody gets stuck if the AI is slow.

There are no background workers: state moves forward when a phone checks in, and every rule is enforced on the server inside database transactions. An end-to-end test drives the whole product over HTTP as a dozen simulated people. It also installs on your phone like an app.

Challenges we ran into

Matching two strangers safely without ever revealing where either of them is: we match on rough distance, meet at the quest spot, and delete locations at the match.
Making the demo work for a judge who's alone, with no partner, no student email and no GPS, without faking the real flow.
Dozens of AI coding sessions working in parallel on one repo. We had to design the workflow (worktrees, small PRs, shared contracts) so they didn't collide, and at one point our merges used up the hosting platform's daily deploy limit.

Accomplishments that we're proud of

A complete loop, from "I'm free" to a match, a quest, AI-checked proof and a recap, live on a real domain with real sign-in and a real database.
A design that feels calm and hand-made rather than like another social feed.
Wellbeing numbers that are private and never ranked against anyone else.

What we learned

Small, safe, structured activities are what make meeting a stranger feel normal: the quest does the work small talk usually does.
Keeping AI in a narrow, checkable role (judge the photo, write the recap) made it reliable enough to demo.
Shipping in small pieces all day beats one big merge at the end.

What's next for Amble

A pilot on the UCalgary campus. Give-back and skill-swap quests with local organizations. Measuring the before-and-after mood check and the familiar-faces count across a term to see whether it really helps.

AI disclosure

Building it: we used Claude Code (Anthropic) as a coding assistant throughout, with many sessions working in parallel on separate branches that we reviewed and merged. It wrote a large share of the code. We made the product decisions: the idea and focus areas, the safety rules, the metrics, the quests, the design direction and what to cut. Every team member can explain the code.
Inside the product: Claude Haiku (Anthropic) checks proof photos and videos and writes recap summaries. It doesn't write the quests: we did, by hand.