Tech Stack that founders use for building AI-Driven SaaS

We've seen how AI has played a significant role in every startup, especially talking about SaaS and Mobile Apps that have some sort of AI features in them.

Most of the time it's directly using the APIs either from OpenAI or some other Generative AI models like Gemini or Sonnet model on AWS. Sometimes even startups train their own AI model to have an in-house solution.

Thus in this article let's try to understand what options a startup has that wants to launch its first MVP in the market. To understand things better we will first check what Open-Based AI Models we have that we need to train and manage their infrastructure and then we will see the Close-Based AI Models from where we can directly plug the APIs in our application.

1. Open-Based AI Models: These types of Models that those where you can directly download an empty model on your servers, and you can train it and blend the model as per your requirements. These types of AI models often take time to become efficient as they require a lot of training data and continuous monitoring, but once they are trained they will reduce your cost and will be more accurate for your product as it's only trained on your dataset.

Some examples of these models are,

1. Llama 3.1 or 2: I've been personally using Llama 2 in one of my projects but recently we got introduced to Llama 3.1 which supports up to 128k context window.

2. Mistral 7b: You can use this model for providing accurate and detailed responses, especially on mathematical problems.

3. Falcon 40b: It's an open-source model that is trained on 40 billion parameters, this will take some time to start at first and it downloads some really large models.

4. Nitro Jan AI: The Open-source technology that we can use for the AI model to tweak and change things like tokes and temperature. It's a desktop application with a nice interface you can set this on your Mac too.

5. XGen-7B: It's an AI model espesally for Research areas by Salesforce.

There are other AI models also like, BLOOM, OPT-175B, and Phi-1 that you can check out on Hyggingface.

To use these AI models you need to host them someplace, you can use your favorite cloud providers like AWS, Azure, or GCP, or if these are not able to provide you the hardware that you need you can also look into places like Paperspace and Runpod.io.

Great now that we've covered Open-Based AI Models let's jump into Close-Based AI Models that we can use.

2. Close-Based AI Models: As the name suggests these are those models where you can directly use their APIs without any training or deployment infrastructure. Some of the examples are,

3. OpenAI APIs

4. Together AI

5. Google Gemini

6. Azure Open AI ( You can also apply for credits in the Microsoft Startup Program)

7. AWS Bedrock

8. Octo AI

These don't require explanations as they all are directly plug and play kind of thing, but in terms of cost, I would recommend you Sign Up for the Microsoft Startup Program from where you will get credits that you can use in your Azure Portal, and inside Azure Portal you can then use Microsoft Azure Open AI Models for Free. This is helpful if you want to go with Open AI and have no idea how much your usage is and it can cost you.

These were the main parts of the AI Models that you can use for your SaaS separate from this talking about the tech stack for backend language that people often use,

1. Python
2. Nest JS
3. Go

Along with the databases like PostgreSQL, Vector, MongoDB, or MySQL. You can easily find these databases on Planetscale for MySQL, Neon for PostgreSQL, and MongoDB.com for MongoDB, Upstash for Vector.

For frontend development, React JS with Vite or Next JS is good for powering your front end with styling libraries like Tailwind CSS, Material UI, or Shadcn/UI.

Now these languages and databases might change as per your SaaS and what features you are building so make sure to check on them which is more fisible for you.

There are some other services also that I've seen SaaS founders using but it's not anything specific to SaaS like, Ably for Sockets and FusionAuth for managing users on the platform.

This was pretty much all, let me know in the comments if you have any doubts or want to share your tech stack I would love to look into them for you.