Integrate DSPy into the form generation #68
Conversation
* auth.py in ./ * added domain_model.puml of the application
- On /new_label a new label is created in the cache. - Workflow diagram documenting how the user will proceed.
* analyze is now a POST request * log the analyzed label * s/test_document_store/test_label
* feat: use pydantic to check the fertilizer extraction json * fix: remove lint flags * fix: eof rule * fix: eof form
There was a problem hiding this comment.
As far as I can tell this looks good.
A few idea crossed my mind when I was looking at the sanity checks. Right now we are doing the bare minimum but it as the fields we are expecting are almost all strings it would be interesting to see if it is possible/worth it to store the numerical values back in numerical format especially when the units are implicit.
I also noticed some fields expect values in certain formats that we might be able to verify with simple pattern matching (e.g. this field expect a percentage so it needs to have a % sign, etc). Although, I don't know if a pattern matching check failing is ground for us rejecting generations or if it would instead be better to flag such inconsistencies in the front-end for human review and let them decide.
Use DSPy signature to structure the prompt.
Roadblocks:
e2e testingwait for the user testingClose #66
Close #67
Close #69 (since the merge of #70)