These are great suggestions. In my particularly usecase I found an easier way to go about it.
I’m working on the GitHub - thmsmlr/instructor_ex: Structured outputs for LLMs in Elixir · GitHub library and the JSONSchema is getting sent into an LLM. What I realized is that the LLM should be resilient to whether the description is at the schema level or the field level. So I decided to just do something like,
defmodule SpamPredicition do
use Ecto.Schema
use Instructor.Validator
@doc """
## Field Descriptions:
- class: Whether or not the email is spam
- reason: A short, less than 10 word rationalization for the classification
- score: A confidence score between 0.0 and 1.0 for the classification
"""
@primary_key false
embedded_schema do
field(:class, Ecto.Enum, values: [:spam, :not_spam])
field(:reason, :string)
field(:score, :float)
end
In some sense, for my usecase, this is even more flexible because you can write whatever you want about the semantics of this schema in the schema level @doc and the AI can make the appropriate associations, since you know.. it’s AI or whatnot.
If I come back to this under other circumstances i’ll definitely explore these solutions. Thanks y’all for your help!
