In my company, we created a UI that uses ChatGPT to create Cypher queries for our knowledge graph. This is fabulous and works very well. The results are shown as tables or can be downloaded immediately as Excel files.
Now, the only drawback is that we don't trust a "public" LLM totally. In the end, we want to run our own backend and our own Language Model.
Are there any recommendations - open source or not - for LLMs that are good at creating CYPHER queries (besides OpenA) and that can be run independendly without OpenAI, Microsoft or AWS in the background?
1 Like
Hi @pingelsan if you can't find a LLM that can't create Cypher to query Neo4j .. you could explore LangChain "procedure calling" feature , where you can code a normal function in python given the results as pandas data frame to your own LLM ..
Just ideas ..greetings
I just remember this fantastic article by
@bratanic_tomaz
Hope this help you and keep us updated. 
1 Like
Can I know how you got the output shown in tables or downloaded as excel files ?
Thanks in advance :)
I was also showing output in tables. I gave few-shot examples such that entities used in all the hops are returned, not just the final destination entity. Then, I parsed the response from Neo4j to create a table with the entity-types as column names, and each row shows all the entities that were "hopped" over to get to the final entity.
Hi Nikhita, haven't been here for a while.
The Cypher queries that the LLM creates usually have some node property in the result set, like customer.name, order.date and things like that. Our Python backend takes these results and creates a table from it which is displayed in the frontend. If someone clicks on "download Excel", an Excel file is created in the backend for download. No magic here... ;-)