Button down the "Getting started" guide

[dglazkov]

Draft here.

Let's rewrite it as a "build a Board Game Card Maker" tutorial: starting with a blank board, write a fun little app that comes up with a board game card based on our drawing. Finished board here

The progression is as follows:

• [ ] Onramp/Getting started (mostly complete)
• [ ] Tour of card maker. Objective: show what we're going to build, explore debugger features, preview, etc. using an existing board.
• [ ] Explore blank board (mostly complete). Objective: get a hang of "hot reload", immediacy. code
• [ ] Add a gemini text node. Objective: show how to add nodes, talk about kits, node factories, ports, etc. code
• [ ] Add a prompt template node. Objective: show how to connect input/output ports of multiple nodes, and use multiple kits. code.
• [ ] Add a random letter node. Objective: introduce code node, what it can and can't do. code
• [ ] Invoke the webcam board. Objective: introduce the concept of remixing and reusing boards using core.invoke node. Talk about the similarity of boards and nodes. code
• [ ] Add flourish. Objective: introduce the concept of input/output Schema and using it to provide hints to UI and form a useful API. code.

Notes:

• Going with gemini.text right away seems right. @dalequark commented that this is probably why people come to try this in the first place, and who we are to keep them waiting?
• Invoking a Webcam board is a pretty interesting trick. The board is pretty simple itself, but it does take care of the need to mess with strings to formulate a valid multi-part Gemini Pro Vision API call : the board has a "prompt" and a "picture" arguments, which is all we need. It also teaches that invoking boards is easy and encouraged.

[PaulKinlan]

Just as a note - We're going to have to also work out the on boarding at the point where you use your first prompt with an LLM, because there are some hurdles outside of our direct control:

• Each hosted LLM requires an account and they are all pretty complex to set up (often requiring a credit card)
• Gemini/PaLM is still not available in the UK (and Europe) - https://ai.google.dev/available_regions
• I imagine we will get interest from developers who only use Mistral or OpenAI, we should have a clear path that doesn't turn them away (recipes/apis/llms/* should help here)

[Mearman]

@PaulKinlan

although they're not as performant, I'm a fan of using your Xenova kit as there are no service dependencies other than loading the model.

[dglazkov]

I am trying to figure out how to best incorporate the Card Maker board into the guide. Maybe rewrite it to be basically the "write from scratch" Card Maker tutorial?

[dglazkov]

The beats of the story:

• blank
• add gemini.text node
• add template
• add markdown output (optional)
• add random letter maker
• add invoking webcam as drawable

[dglazkov]

The gist containing code progression is here: https://gist.github.com/dglazkov/cfdd753564b84bbcc83c30b022c6fecb