Ask HN: How are you building apps that use LLMs?

13 points by jrvarela56 a year ago

Iterating workflows such as the ones described here https://github.com/openai/openai-cookbook/blob/main/techniqu... is a bit tedious with Jupyter.

I've been using https://natto.dev/ to try out different workflows side-by-side, but was curious to see what other HN users were up to.

I'm looking for environments (Jupyter, Natto) more so than libraries like https://github.com/hwchase17/langchain or https://github.com/jerryjliu/gpt_index.

jrvarela56 a year ago

As I mention briefly in the description, https://natto.dev/ has been awesome. A few things you can do:

- Avoid auto-running certain panes

- Split paths: try different approaches with the same input and see them side-by-side

- Collaborate: you can share canvases with others

- Create UIs: panes can contain JSX so you can build parts of the UI next to your workflows

Given the strong AI tooling ecosystem, it would be awesome to have this kind of environment for Python.

  • zainhoda a year ago

    I’m working on something like this for Python. Live code editing, forking other users code, and built-in Web UIs that let your end users interact with your prototype. We’ll be ready next week for a handful (probably around 5) initial test users if you’re interested.

    • jrvarela56 a year ago

      Sure thing! Send me an email (see profile)