Can Vectara’s capabilities support some kind of product search? If I have an inventory of products for an e-commerce site, with each product containing a title, description and an exhaustive list of descriptive keywords — could I use Vectara to power natural language search? (e.g. “show me spacious L shaped living room sofas that would look good in a modern home”)
If so, what would be the most optimal way to structure this inventory data?
Vectara can absolutely help with product search use cases – that’s one of the core use cases we see from our users.
In terms of data modeling, here’s what we generally recommend:
1 document per product
Use a section for each search feature you’d expect a user to have full text search on. For example in your case, I’d add title, description, and keywords each as separate sections in Vectara’s terminology. These are essentially providing hints to Vectara on what you want to be used for natural language matching
Add structured metadata to the metadata_json. For example, if you wanted to provide a user with the ability to hard filter by price as well, put a price field there. If you have inventory that you want to allow users to do things like “only show me instock” that would be a good place to put that
If you’d like me to connect you with someone at Vectara that can help further and/or let you know some of the more advanced functions of the platform that aren’t in the Growth plan, don’t hesitate to ping me! shane@vectara.com