Preciazo
Preciazo
The lowest price, no fine print.
Platform
Web responsive
Brief
Preciazo is a multi-store price comparison tool and price tracker built for the Mexican digital shopper. Its mission is to help users find the real lowest price and give them the certainty that deals are legitimate by checking them against a product's price history — especially during events like Buen Fin or Hot Sale. By making daily item tracking completely free and requiring no forced sign-ups, the platform democratizes smart shopping in Mexico.
Inspired by the honesty of Mexican 'hard discount' retail graphics (think Tiendas 3B or Dr. Simi) and the frankness of traditional street-market signage, Preciazo rejects the polished aesthetic of foreign corporate software — no glossy gradients, no frosted glass. Instead it embraces the idea of 'The Honest Poster' (El Cartelón Honesto): firm black-ink borders, flat paper-colored backgrounds, hard blur-free shadows, and a strict hierarchy where the price and the buy button are the undisputed visual heroes. The design is strictly mobile-first, built for fast interaction from a phone.
El Cartelón — signature price panel
The lead module on the product page. It concentrates the main price, the redirect buttons, and the tracking option on a bright yellow block that evokes physical sale tags.
Cut-paper physics
A tactile interface where hard shadows carry no blur. On interaction, CTAs lift on hover and physically sink on click, simulating the three-dimensionality of cut paper.
Typography with voice and force
The robust, condensed Anton typeface gives maximum volume to prices, complemented by Archivo for friendly, sentence-case copy.
Frictionless mobile-first flows
Designed for the Mexican shopper's mobile use, with generous touch targets (minimum 44px) and processes simplified so tracking starts instantly.
Technology
A hybrid architecture that pairs Next.js interactivity on the frontend with independent Node.js and Python microservices to guarantee scalability, agile price crawling, and cost efficiency.
- Next.js
- React
- TypeScript
- Recharts
- Node.js
- Express
- Python
- FastAPI
- Scrapling
- PostgreSQL
- Google Vertex AI
- Gemini 2.0 Flash
- Google Cloud Run
- Cloud SQL
- Resend
- PostHog
What we built
Multi-store comparison
Search any product and instantly compare its price across popular Mexican retailers (Amazon, Mercado Libre, Walmart, Liverpool, Coppel, and more) in one place.
Historical price tracking
Clear charts of recorded minimum, maximum, and average prices, letting shoppers verify whether a discount is real or the price was previously inflated.
Personalized email alerts
Set a target price — as a discount percentage or an exact peso amount — and get an automatic notification the moment the product hits your goal.
Deferred sign-up with magic link
Start tracking a product optimistically with just your email — no passwords, no upfront friction.
Intelligence at the core
AI product matching
Automatically identifies and groups identical products across stores even when their titles and photos differ. It generates vector embeddings with Vertex AI (text-embedding-004) and computes similarity in PostgreSQL (pgvector), then uses Gemini 2.5 Flash to tell price variants apart — like different storage capacities.
Autonomous data extractor
A Gemini-based agent with browsing tools (function calling) pulls structured data — title, price, brand, images — straight from a retailer's DOM or JSON-LD. It removes the need for brittle CSS-selector scrapers that break with every store redesign.
AI self-discovery search
No more dead-end 'no results' pages. When a product isn't in the local catalog, a background agent searches supported retailers in real time, validates and imports the new products, and streams them to the user (via SSE) in seconds.
What we solved
Matching products with no shared SKU
Every e-commerce site lists the same item with different titles, photos, and SKUs. We built a pipeline that generates vector embeddings and computes their similarity in the relational database, then validates variants (like storage capacities) with language models to group them under a single 'master product'.
Cost-effective data extraction against anti-bot defenses
Daily price monitoring on heavily protected sites like Amazon MX and Mercado Libre is often blocked from the cloud. We solved it with a Python microservice optimized with Scrapling to emulate real residential traffic, plus dynamic CSS selectors that cut scraping to just 4 seconds — lowering paid-proxy costs.
Removing the upfront registration wall
Requiring a traditional password account before setting up alerts caused users to drop off at the key point in the funnel. We solved it with deferred sign-up: the system optimistically creates a session and starts tracking with only an email, holding real alert delivery until the user confirms their inbox via a secure link.