Feature: AI project keyword suggestions

[van-go]

Overview:

• Adds view that connects to our chromadb instance, retrieves related projects, and connects to an ollama model running on Tejas to retrieve a list of suggested keywords for a given project, provided the project title and description.
• Queries to the api are rate limited by the client to 1 per second.
• Keyword suggestions component that is rendered in the project creation modal, project edit modal, and perhaps the project publication pipeline
• Adds hook to retrieve suggestions from endpoint, only if project title and description both exist

PR Status:

• Ready.

Testing Steps:

1. Update your designsafe.env secrets locally with the latest from stache, which include secrets for chromadb and Tejas.
2. Edit a project, and ensure it has a project type set
3. Edit your title and description, and confirm you see relevant keywords suggested
   • This should work for EVERY project type: Other, Experimental, Field Research, Simulation, Hybrid Simulation, and Research Software
   • if either field is empty, you should be prompted to "Enter a project title and description to see keyword suggestions."
   • If both are filled, but then deleted, you should get the same prompt
   • When new title and description are entered, you should see "Finding suggestions..." until suggestions appear.
4. The above is true at the entity level
   • Simulation --> Add Simulation
   • Experimental --> Add Experiment
   • Hybrid Sim --> Add Hybrid Sim
   • Field Research --> Add Mission
   • Field Research --> Add Document

UI Photos:

TODO:

Along with adding Natural Hazard type at the project level to the RAG, each entities has their own required fields that can be helpful, such as:
Simulation --> Simulation type, and Hyb Sim --> Hyb Sim type. None of the other ones have required fields that might be helpful.

[fnets]

This all looks great to me! Good job!

[fnets]

The linting check is failing for designsafe/apps/api/ai_keywords/views.py

[fnets]

This is a great idea

[fnets]

LGTM and works locally! Also super clean, thanks! I had some feedback, but feel free to disregard it if it doesn't make sense to address.

[fnets]

Maybe I'm opening a can of worms here, but is this how we want to handle errors? Or should we let the user know that there's an issue with the model? Or should we log the error for our own debugging?

[rstijerina]

I think if we encounter errors for this feature, they should be hidden from the user. We log errors on the backend, so good there.

[fnets]

Another annoying can of worms that I might be opening up, but are there any CSS classes or styling we would want to use instead of setting it manually in the JSX like you do in this chunk? We haven't talked to Wes or anyone yet, so maybe that will be a to-do for later.