Startups, Saas & Financing and API/Hosting-costs

Part 2 2022/23
1

Hi everyone.
I imagine several of you have thought about creating some side project that is more than a proof of concept. But in the event that the project would take off and acquire lots of users, you would need to consider scaling the infrastructure to support lots of traffic/users. And if your solution relies on calling third party API’s that cost money or doing heavy GPU-needing stuff, you would need to take that into account in deciding how you finance / take revenue for your service. Unless you go with a “make some losses and see what happens” approach :slight_smile: .
If your solution is compute heavy per session, you would need to spin up parallell instances of your solution.

I would like to start a discussion about profitability and feasibility of side projects. What are your thoughts, what are the risks, what are the opportunities, what tools do you recommend, say for

  • frontend hosting
  • compute/api server
  • model storing
  • fine-tuning based on new data during coming in from users
  • payment system
  • subscription models
3 Likes
2

Hugginfaces inference API could be used for servicing the data via API’s
You would need to calculate the cost, to see what kind of financing model you need.

There are several services you could use for frontend hosting if you for instance go with next.js and react.

For instance Vercel you could use as the frontend in the beginning. It understands next.js projects out of the box. and has a free tier for personal / non-commercial projects

1 Like
3

This is a good comparison for choosing frontend hosting Heroku vs Netfliy vs Vercel vs GitHub Pages vs Firebase vs Vercel - Ritza Articles

1 Like
4

Here is a more in depth comparison of netlify vs vercel

Both have a free tier, but the vercel version does not let you use to for commercial use.

5

Did you find a good way to send confirmation/password reminder emails?

6

I am investigating how to do as low cost as possible in the beginning stage of a project. Therefore, i will not at this time go down the rout of doing inference for my project in the cloud, but i will create a queue system and my own machine will pull the jobs from the queue and upload the results to a db in the cloud (in this case firebase firestore) and the user will just see the results when they have been processed.

This way, i do not need to pay for a GPU machine that sits idle most of the time.
My own deskotop with an GeForce RTX 4090 and 64Gb ram, can take the jobs when they appear and upload the results.

So the system will have 3 parts

  • the frontend
  • the api to create queues and deliver data when it is available.
  • job processor

It is a bit silly but it is a proof of concept. At least it removes the cold start problem if you shut down a server, and it also prevents you from having to pay for idle time.

1 Like