Compensa?
A Portuguese-language price comparison site. It compares products by unit price (per gram of protein, per 100 g of creatine, per ml, per diaper, etc.) across brand stores, pharmacies and supermarkets:
- Supplements: whey, creatine, protein bars, omega 3, peanut butter
- Skin care: sunscreen, moisturizer, vitamin C
- Home and baby: diapers, toilet paper, laundry detergent
- Pet food
Preview locally
make serve
Open http://localhost:8001. Category pages (e.g. /whey, /racao) are served from index.html.
Update the catalog
npm install
npx playwright install chromium
make update-products
npm run check
SOURCE=<brand>, SOURCE=<category key> (e.g. petFood) or SOURCE=<store> (e.g. Cobasi) refreshes only part of the catalog. Results go to data/products.json and assets/products/.
- Whey and creatine: brand stores (Growth, Max Titanium, Essential, Probiótica, Nutrify, True Source, Soldiers, Dux, Adaptogen, New Nutrition, and creatine-only stores), scraped with Playwright or read from VTEX and Shopify APIs. Nutrition tables, ingredients and allergens come from the product pages; Portuguese OCR (Tesseract) is used when the table is only an image, checked against the label's energy value.
- Sunscreen: Nutriex, Adcos, Dermage and Drogaria São Paulo, with store reviews keyword-counted per trait (stinging, sticky, oily, white cast…).
- Other categories: VTEX store catalogs (Drogaria São Paulo, Pague Menos, Savegnago, Zona Sul, Cobasi, Adcos, Dermage and the supplement brand stores).
- Mercado Livre (sunscreen only): catalog attributes (water resistant, oil free, tinted…), matched by barcode. Needs
ML_CLIENT_ID and ML_CLIENT_SECRET in .env; without them the previous data is kept.
Run npm test after changing the OCR parser, review keywords or label parsing.
Publishing
make build copies only the public site (HTML, CSS, JS, catalog, images, category pages and sitemap) into dist/. Scraper scripts and OCR data are not published.
