Day 88!

Struggled yesterday and this morning to get my terrible MVP online. Actually, it’s not even an MVP. It’s really a proof of concept, if barely that.

But Y Combinator says if it’s not shit, you waited too long to ship. So F it.

Tried sticking it up on Vercel’s python runtime through a basic Flask app, but should’ve realized that the limits and timeouts relating to a serverless function architecture wouldn’t be the easiest way to get it up and running.

And we’re moving fast. So what do we do? I have (and pay for) a Django app running on heroku, so let’s stick it on there (despite having not touched that terrible cranky codebase I built within a month of learning how to code and is the textbook definition of bad practices and technical debt).

So I quickly ported over my Flask code to Django (thanks, ChatGPT!), figured out what else I had to stick in the requirements.txt, and tried to deploy!

WOOPS

Including things like huggingface and pytorch made it WAY TOO BIG to deploy given heroku’s slug size limitations (also - not at all what I normally think of as a slug in web dev, so that’s confusing).

Some troubleshooting and stack-overflowing later, and found out that including the cpu-only versions of torch and torchvision would probably solve the problem and it did! Ish. The app builds, although it’s still over the ‘recommended slug size’.

And then - production server crashed. Rolled back, checked out the logs, mis-spelled some of packages in the new view’s import statements. Crap. Fixed, deployed - works!

Ish. Running any queries in that view immediately maxed out heroku’s memory limitations (are we seeing a theme here? No matter, moving fast!). Jacked up my spend to very high performing dynos I would never otherwise need, and it mostly works without crashing! This is painful.

Now, users!

Called up four lawyer users of my current app, asked them to try it out. Three answered, two had time to try it out right there on the spot - clear and immediate feedback was “Yes, this is good and if it worked well I would absolutely pay real money for this”.

Okay! 72 hours in, deployed a proof of concept and got some positive feedback.

What’s next?