Seeking Help to Implement Offline Vector Search in Alteryx

Question

Hello Alteryx Community,

I'm hoping someone can kindly help me with this scenario. I've built a vector search index using Google Cloud's Vertex AI, and it works well, but the costs are adding up, so I'd like to find a way to run it offline in Alteryx.

Here's what I've done so far:

* I created the vector embeddings and deployed the index in Vertex AI using Python code in a Jupyter notebook.
* The resulting vectorized data is in a JSON file (it's JSONL format with fields like 'id', 'title', 'embedding' (dense vector), and 'dimensions' (sparse indices)).

My question: Is there a way to replicate this vector search functionality offline in Alteryx? For example, using the Python Tool with libraries like Faiss for dense search and something for sparse/hybrid queries? Since I already have the JSON file with the vectors, I'd love to load it, create a local index, and run queries without relying on Google Cloud.

Any guidance, sample workflows, or code snippets would be greatly appreciated! I'm open to suggestions on handling the dense and sparse embeddings in Alteryx.

Thanks in advance for your help!

hybrid-search-v1.0.0.zip

apathetichell · Answer

@buddhiDB --- This is a weird post. The short is 'yes' --- but how you get to 'yes' is your battle - and Alteryx doesn't have a role here. Maybe it's like a last mile trigger or something but it's basically meaningless.

You have vector embeddings. is this a .bin file? if so you run FAISS locally. You can set up a docker container and expose it locally. you can set up whatever type of server (baremetal/cloud/container/container on cloud/container local/container on rasberry pi) you want --- and hit it and self host your embeddings. Your problem --- you cannot create new embeddings without an llm. Can you self-host an llm? Sure. Maybe look at HuggingFace for ideas on how to do that. Are you a google shop? You can expose this all on cloud run via self-hosted containers.

How Alteryx plays into this? You will need to build out an api and have Alteryx hit this like any other api --- or have Alteryx excute python code like any other python code.

dreldrel · Answer

You want to run it offline, then you'd need to have an offline LLM model in your local system.