The Tech Stack for your Modern day AI Product Engineer

The tech stack you choose when starting a project creates a potential path for you. Every piece of the stack has its trade-offs, and when shipping certain products their are certain trade-offs you should be more willing to deal with than others.

In choosing my stack for AI products, I look for three things:

• Familiarity. I like to use tools and languages I am already familiar with so that I can build quickly.

• Good support/Big Community. Good support and a big community mean you have more references and documentation to refer to which aids up development time, especially if you are a solo developer.

• Easy to iterate with. I like tools that allow me to quickly change up styling, auth flows, and manage state.

With that being said, these are the tools that I am currently using to ship products for clients and to the community in the AI space.

Next.js

Next.js is more or less the standard production React framework. It has a huge community, an amazing developer experience, and the company behind it, Vercel, has released a whole AI SDK that makes it very easy to get up and running with your favorite language models using their pre built templates. You could have a whole gpt-wrapper application up and running in 5 minutes using their pre configured templates.

Tailwind CSS

Tailwind is an opinionated styling library that allows you to use utility-first classes directly in your markup to apply your design. Again, it has a huge community, a ton of free and premium templates available to ship products quickly, and it is easy to iterate on if the styles need to be adjusted. It also comes with some neat tools like a purger that automatically gets rid of styles you don't use, so if you write sloppy css (like me :p), then it's perfect for you!

Typescript

Coding like a cowboy with JS is great and all, but it can lead to hours spent debugging type errors and retracing call stacks in the browser. I like to avoid all that by using Typescript. Sure you take a bit of extra time during build, but I find the extra time spent during build is always less than the time I would have spent debugging the application if I had used vanilla JS.

Clerk

Clerk makes it very easy to setup Auth flows, so it is my choice for AUTH even when using other backend providers. I think the developer experience is unmatched.

Supabase

Supabase is awesome in that it's a backend as a service, so it can become a drop in replacement for alot of tools that people piece together from many providers. They also have an awesome, open source, and cheap vector database that is great for LLM applications. Coupled together with something like Langchain, you can build AI agents that have access to a real time vector database.

Langchain

An AI specific framework, Langchain is a great tool that allows developers to build context-aware LLM powered applications. Langchain can provide LLM's access to a variety of different tools and API's, and they offer some integrated development products to streamline getting those applications into production.

LangSmith

An offering of the Langchain team, this gives developers observability over their application and logic chains and can provide advanced debugging, evaluating, and testing tools specifically designed for AI applications.

Replicate

Replicate is an amazing tool that allows founders to build applications around a number of open source models, all with a unified API. It makes creating applications that make use of multiple models easy and affordable. Its API docs are very user friendly and well supported.

OpenAI API (of course)

I put this one last because this is what everybody knows. GPT-4, DALLE-3, and the assistants API have some of the best, if not the best, results when it comes to generative AI and LLM's. Their API is very well documented and easy to use, a huge community exists that are constantly putting out templates and well organized guides, and well, its OpenAI. They are definitely leading the charge in many ways and some of the best products I have personally built for clients have been built around their API.

This list is not all encompassing. Their are so many tools out there it would be impossible to fit them all in one post. These tools for me, however, have provided me the ability to ship quickly, often, and have allowed me to iterate on them based on user feedback.

What tools are y'all using to ship your AI products this year?