I've been loving building with AI, and over the past few years I've been leaning more and more into Typescript (and bun).
My team at inference.net is constantly trying to get more leverage out of AI and find ways to setup our codebase to be able to increase the level of correctness that our AI is able to write code at.
This starter repo is a very opinionated way to lay out a repo to lean into AI heavily.
It leverages Cloudflare Workers as a deployment target for the API (my goal is to never have to deploy an API on a AWS/Azure/GCP server ever again unless I get to a scale that absolutely requires it).
I'm also a big fan of tRPC for its type-coupling between the backend and frontend (you can also use Next App Router to get the same benefits).
Some key properties of the project:
- Its a monorepo, so everything the AI needs to see is in one place
- Very heavily leans into Taskfile (similar to Makefiles) to give the AI commands it can run to do complex tasks that give it feedback on its code (like linting/formatting/running the project/generating migrations)
- Uses strict typing to give the AI a strong feedback loop on the code it writes
- Uses the repository pattern to be able to easily swap out implementations for persistence
- Uses bun for its ease of use
- Leverages DI to allow configuration to be injected so it can easily be overridden when deploying to Cloudfalre Workers
- Has a very opinionated way to leverage markdown documents to give the AI strong context on how and when to do things
- Can be run end-to-end locally with zero external dependencies
*I'm looking for feedback.* Please don't hold back, tell me where you think it can be simpler (one area is maybe swapping out GoTrue for better-auth). And if there are patterns that you use to get more leverage and correctness out of AI, I'd love to learn about them!
Thank you for checking it out :)
- abe