Totes-not-amazon.com: Markov Chain Generation for AWS Announcements

I challenge you to reliably tell the difference between AWS and Totes-not-amazon.

The former, of course, is AWS’s announcement blog. It’s capably written, but AWS’s word-soup product names and features make it sometimes sound like it was written by a script. So I built that. Meet https://totes-not-amazon.com! I set it up to reroute you to a reusable link so people can share particularly funny results. Click nearly any link to generate a new post.

Implementation-wise, it goes something like:

  1. An offline script scrapes all 3k+ aws posts under aws.amazon.com/about-aws/whats-new and produces a json dump of these. This is in ruby because I just wanted to get this step done fast and that’s the language I know best at the moment.
  2. Another offline script takes that json dump of blog posts + titles and fills up a markov model for text generation. I used the excellent Markovify libaray for that. I went with python for this script because it seemed to have good markov libraries and also I needed to share some logic with step 4. After generating the markov model, this script dumps that model to json.
  3. Now, online, when a user hits totes-not-amazon.com/, the frontent js hits an api backed by AWS API Gateway, which triggers a lambda function.
  4. The lambda function (which includes a copy of the dumped markov model json) loads up the markov model and generates a new randomized blog post + title, then returns it to the frontend JS. I’d have used ruby here, but AWS Lambda doesn’t support that yet. I’d have used node, but I wanted to be able to specify a seed for the randomness used in the markov model, allowing me to reproduce especially funny results by specifying a seed. JS doesn’t allow that but python does.
  5. Back in the clientside JS, we get the result of the api call (a blog post + title) and insert it into the page.

For the lambda function, I used the Serverless Framework for the first time, which was a pretty nice way of managing a lambda function.

For hosting + deploying the static files I used S3 + Cloudfront + ACM + Route53 (through my own Scarr tool).

If you want to see the code, it’s kind of a mess, but it’s at github.com/kkuchta/aws_markov. For a silly one-off like this I’m unlikely to go back and clean it up unless someone really wants me too.

Anyway, this is just some silliness and an excudes to mess with a bunch of random tools + languages. Nothing serious today. Go play with it and tweet your favorites to @kkuchta!