Tracking supermarket prices with Playwright

How does the ipv6 rotation work in this flow?

Nice writeup. I've been through similar problems that you have with my contact lens price comparison website https://lenspricer.com/ that I run in ~30 countries. I have found, like you, that websites changing their HTML is a pain.

One of my biggest hurdles initially was matching products across 100+ websites. Even though you think a product has a unique name, everyone puts their own twist on it. Most can be handled with regexes, but I had to manually map many of these (I used AI for some of it, but had to manually verify all of it).

I've found that building the scrapers and infrastructure is somewhat the easy part. The hard part is maintaining all of the scrapers and figuring out if when a product disappears from a site, is that because my scraper has an error, is it my scraper being blocked, did the site make a change, was the site randomly down for maintenance when I scraped it etc.

A fun project, but challenging at times, and annoying problems to fix.

Isn’t this a use-case where LLMs could really help?

Yeah it is to some degree. I tried to use it as much as possible, but there's always those annoying edge cases that makes me not trust the results and I have to check everything, and it ended up being faster just building some simple UI where I can easily classify the name myself.

Part of the problem is simply due to bad data from the websites. Just as an example - there's a 2-week contact lens called "Acuvue Oasys". And there's a completely different 1-day contact lens called "Acuvue Oasys 1-Day". Some sites have been bad at writing this properly, so both variants may be called "Acuvue Oasys" (or close to it), and the way to distinguish them is to look at the image to see which actual lens they mean, look at the price etc.

It's true that this could probably also be handled by AI, but in the end, classifying the lenses takes like 1-2% of the time it takes to make a scraper for a website so I found it was not worth trying to build a very good LLM classifier for this.

Unfortunately no, but i can share some insights that i hope can be of value:

- Tech: Everything is hosted in AWS. We are using Golang in docker containers that does the scraping. They run on ECS Fargate spots when needed using cronjob. The scraping result is stored as a parquet in S3 and processed in our RDS Postgresql. We need to be creative and have some methods to identify that a particular product A in store 1 is the same as product A in store 2 so they are mapped together. Sometimes it needs to be verified manually. The data that are of interest for the user/site is indexed into an Elastic search.

Things that might be of interest: - We always try to avoid parsing the HTML but instead calling the sites APIs directly to reduce scraping time. We also try to scrape the category listing to access multiple prices by one request, this can reduce the total requests from over 100 000 to maybe less than 1000 requests.

- We also try to avoid scraping the sites during peak times and respect their robots.txt. We add some delay to each request. The scrapes are often done during night/early morning.

- The main challenge is that stores can redesign or modify which make our scrapers fail, so we need to be fast and adopt to the new changes.

- Another major hidden challenge is that the stores have different prices for the same product depending on your zip code, so we have our ways of identifying the stores different warehouses, what zip codes belong to a specific warehouse and do a scrape for that warehouse. So a store might have 5 warehouses, so we need to scrape it 5 times with different zip codes

There is much more but i hope that gave you some insights of challenges and some solutions!