I Know a Place

Recommendations from people you actually trust — not a scraped review, not an algorithm's guess.

Every place on this site was recommended by someone real, with their own rating and their own words. It's a browsable map + list for exploring Toronto through 56 recommendations from 22 friends, plus a "Plan my trip" tool that builds a real day plan out of them (time / energy / budget in, a few realistic plans out — no AI guessing, no generic itinerary).

Live site: iknowaplace-eta.vercel.app

Why this exists

Full story is on the site itself (the "Why this exists" link), but in short: recommendations from friends are usually scattered across texts and group chats, disconnected from how much time or money you actually have. This turns them into something usable.

Stack

Plain HTML, CSS, and JavaScript — no framework, no build step, no backend. index.html holds the markup and the logic, css/app.css holds the design system, and the content lives in two JSON files in data/. Nothing is compiled, so any static host (this one runs on Vercel) can serve the folder as-is.

The one library is the map: MapLibre GL JS draws OpenFreeMap's free vector tiles — no API key, nothing self-hosted. The map's look is a plain style file, map/ikap-atlas.json, that can be edited directly or in Maputnik.

Run it locally

The page loads its data with fetch(), and browsers block that on file:// URLs — so opening index.html by double-clicking it won't work. Serve the folder over http instead. With Node installed, run this in the project folder and open the address it prints:

npx serve

Data

Sourced from a Google Form filled out by real friends, then cleaned and anonymized (first-name aliases only, no last names). It lives in two files you can open and edit directly:

• data/places.json — the 39 places (name, category, neighbourhood, coordinates and how confident they are, address, rating summary)
• data/recommendations.json — the 56 recommendations (who suggested it, their rating, their note)

The two are linked by place_key: every recommendation names the place it belongs to. Those keys are permanent — never renumber them.

Design

The look has a written system behind it rather than a set of ad-hoc choices — palette with measured contrast ratios, two typefaces with one job each, how the map is drawn, and the rules the components follow:

• docs/DESIGN_SYSTEM.md — the system as built
• docs/VISUAL_REDESIGN_PLAN.md — how it got there, and what was tried and rejected
• docs/REAL_MAP_PLAN.md — replacing the drawn map with a real one

Status

Prototype — first city is Toronto. Built as a hands-on project to learn AI-assisted ("vibe") coding from the ground up.