Summary
ChatGPT plugin, which could also run standalone on a website, that lets the users ask the german law code (as scraped from gesetze-im-internet.de) for information.
Motivation
- consider that it could support languages other than german; in particular Polish, Turkish, English and other relevant minority languages
- does translation to simple language
Questions for LangDock
- the flow part
- anything else?
Log
do I need a backend codebase?no, use supabase edge functions
docker doesnt work, reinstall
- testing edge functions, including such with langchain
- ⭐ following https://github.com/langchain-ai/langchain-template-supabase and https://www.youtube.com/watch?v=29p8kIqyU_Y
- openai key is missing
- two possible reasons:
- not there as envvar (especially not in deployment)
- also
Flow
option 1: “Was passiert wenn man was klaut?” → chatgpt makes some decisions about what question to ask to a plain apt → api responds → chatgpt interprets the answer
option 2: “Was passiert wenn man was klaut?” → the question goes into an api endpoint directly → the endpoint itself uses langchain + embeddings to identify relevant articles → fetches the information and interprets it
Really looks like option 2 is the right one, but maybe there’s option 3 or something inbetween
Supabase + edge functions
Feedback for langdock
- I didn’t have an API to start with, I had to make it and felt like I lacked some info
- your onboarding assumed I already have everything prepared, which got in the way for me
Format
as of 16.05
- some summaries seem reasonable, and there’s a large pole do popisu
- still, focus on closing the loop before improving things
- this seems to lose some information, check how it is based on https://www.gesetze-im-internet.de/bgb
- lots of levels, probably not 100% consistent across all books
yeah
look at github scrapesetze
- nice
make a json schema
- chapter
- which_book
- titel
- untertitel
- kapitel
- “name”
- content
msg:
hey, I want to make a simple API