Hello all,
I currently have a table of more than 1000 rows. I would like to add an ability to the app that allows the user to input natural language to find the top most matching products. For example, “the client would like x for 5 nights in y.” I was thinking of tagging all the rows, but I feel like this would be heavy on AI usage? Is there any better way to create this concept?
Thanks
Hey Matthew, what’s your use case?
Cuz Glide is build for structured data. Normally natural language search is used when information is not structured. So I don’t think they’ll add features that makes it easier to handle messy/unstructured data.
It would be for something like this: https://www.youtube.com/watch?v=zAIb3QZGy4g
It looks cool and maybe for some use cases it’s the best option, but it seems a helper table would be:
See this vid at 2:05
Yeah I could definitely look down that path as I’m sure that will also be useful.
I am trying to explore all AI (feasible) options first and was hoping to gain an insight on how Low Code Agency managed to create this to see if this is something that may be possible.
Thanks for the help 
I don’t know all the details, but I kind of remember them talking about it in the past. I think there is some process where the data is sent to a third party vector database, and then queries are directed towards that database. I’m assuming it returns a lists of Row IDs that can then be used to to query the local tables.
Interesting, I’ll see what I can find
Search the forum for Pinecone. There are a handful of threads about it.
I’ve had a look at Pinecone and this seems to be suitable for my use case however it seems quite complicated so may need to get my head around how it works 
Matthew, could you share what you’ve found complicated so far: is it understanding Pinecone, setting up an API or integration with Glide, setting things up in a Glide app?
I’ve never looked into this topic but I’m interested in whatever insights you can share.
I am the creator of that. AI was used to generate a JavaScript function that acts on data and create values that I can used to return search results. So the cost is only on the AI part to generate the function.
It’s hard to get it right though, a lot of back and forth checks to get to a workable prompt.
Later on, we use Pinecone for other projects, it’s good when you have a set of documents to query data from.
Do you use Pinecone in the context of a Glide x Pinecone integration, or do you use Pinecone independently?
I think it’s just learning a whole new system & concept which has presented a steep learning curve. I have little to no knowledge of vectorised databases. I have managed to create an index and link it to the Glide intergration but I am currently troubleshooting an error I am recieving.
I think right now it is a case of trial and error for me as I am not 100% sure how the Glide intergration functions and what input it expects etc.
If I find out anything new on this topic I will share it here.
Ah cool. I did start to go down the Javascript path but with 1K rows I was thinking it may not be the best option haha
We used Glide x Pinecone.
It doesn’t need to process 1k rows. You just need to add your “input query” in a separate helper table with 1 row, then generate the JavaScript function using an AI column in that helper table (so it processed one).
Cast the function to the 1k rows table using a single value column, and then use a JavaScript function to evaluate that function. In my case, I return a boolean.
You have to be very specific on things like keywords, numbers etc to feed into the JavaScript function so it returns a boolean.
After testing around with Pinecone for a while, this seems to be fitting for my use case! Thank you all for the suggestions.
This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.