realdeal

front page
common pricing tricks
price scan
catalog of stores
location-based scanning

Inspiration

Groceries keep getting more expensive, especially in Vancouver, and stores fill the shelves with "sales" that often aren't real savings. We are all aware of this classic marketing gimmick yet we have no tools to fight back. Thats why we built RealDeal, and app that allows anyone to check if a deal is real before they buy.

What it does

Scan a shelf tag or type in a price, and RealDeal tells you if it's a Steal, Good deal, Normal price or Overpriced, along with how much you save or overpay. It compares the unit price to the 13-week average across Save-On-Foods, No Frills, Walmart, T&T and Loblaws.

It also flags common pricing tricks:

Forever sale: on sale 6+ of the last 12 weeks, so the sale price is really the regular price.
Fake "was" price: the struck-out price was rarely charged.
Multi-buy trap: "2 for $X" barely saves anything.
Shrinkflation: a smaller package at a higher unit price.

Each item also has a 26-week price chart and prices at nearby stores.

How we built it

Frontend: React, TypeScript, Tailwind and D3
Backend: Hono, hosted on our own PC through a Cloudflare Tunnel
Verdict engine: shared TypeScript, so the browser and the server always agree
Scanning: Gemini vision reads the tag and returns structured JSON
Data: real price history from Project Hammer for about 130 items, with generated data filling the gaps

We split the work into about 45 small specs, gave them to AI coding agents in parallel, and used Claude Code to merge the results. We made tweaks and improvements to the project with Claude and Gemini.

Challenges we ran into

Our biggest challenge was collecting enough accurate price data. Stores don't publish their price history, and scraping their sites reliably wasn't possible in a weekend. Project Hammer had real prices, but product names differ between stores, so we had to write strict matching rules and fix many matches by hand. We also had to fix unit mix-ups (per lb vs per kg) and stop Gemini from inventing "was" prices.

Accomplishments that we're proud of

You can photograph a real shelf tag and get a verdict with the evidence behind it in seconds. The app also catches how a price is presented, not just how high it is. It actually works and provides useful information. Try it yourself!

What we learned

Most fake deals come from playing with package sizes or "was" prices, so unit price matters most. Good data is the hardest part of this kind of app. And AI output always has to be checked.

What's next for realdeal

Live scrapers for weekly price updates
Price data contributed by shoppers for each store
Expanding store selection to include stores that are always on sale, like Canadian Tire and Old Navy
Price alerts