[Improve Existing Best Practice Guide]: Guidance on choosing a documentation hosting solution

[riverma]

Checked for duplicates

Yes - I've already checked

Best Practice Guide

Documentation

Best Practice Guide Sections

Trade Studies

Describe the improvement

We've gotten input from folks that they have a need for choosing a best documentation hosting solution, as well as how to get started with it. Currently, we have a table of tools to choose from but its fairly bare bones and difficult to interpret. Let's make this easier: https://nasa-ammos.github.io/slim/documentation/trade-studies/

[riverma]

+1'd by @ramesh-maddegoda, @drewm-jpl, @stirlingalgermissen, @AaronPlave, @nttoole, @rtapella, @MJJoyce, @galenatjpl, @jeffreypon, @Scotchester, @kgrimes2, @pymonger, @hookhua

[jl-0]

Seems like modernizing the VICAR documentation from https://www-mipl.jpl.nasa.gov/mipex.html would fit well with this. It would be ideal to find a hosting source for the public documentation which is not a self-managed server.

[jl-0]

Can't seem to find where on the airflow documentation site I found this, but they have created a channel that was trained on all their documentation so that if you have a paid account you can query the documentation.

https://chat.openai.com › g-lp2HCBHUY-ask-airflow

[riverma]

Nice - thanks for the heads up @jl-0. If you come across other ideas on LLM integration (especially free) please keep us posted!

[PaulMRamirez]

Consider using an open source LLM to create an “Ask ”. For example an “Ask Aerie” or “Ask MMGIS”. This could be built off of public documentation. An example of how to approach this is described here https://www.linkedin.com/pulse/how-build-rag-chatbot-using-ollama-serve-llms-locally-sri-laxmi-beapc.

Unlike the example you’d want to store the trained data in a data store. Testing to see how much documentation can be loaded would be key. Lang chain seems to have support for that. Additionally you’d want to have the “Ask Product” seeded with the documentation so it would not have to load it on each request.

This approach could also work for something that needs to be hosted and trained with internal information to an organization.

Once trained the “Ask ” could be integrated into the docs site for a product and their community slack channel for the product. The former would allow for quick responses to those with common questions.

Suggest using Aerie as a use case. Their public documentation is here https://nasa-ammos.github.io/aerie-docs/ with the source here https://github.com/NASA-AMMOS/aerie-docs.