TL;DR: I’m doing a CatBench Vector Search on AlloyDB webinar (hacking session style!) with Google on Wednesday, July 23th at 9am EDT.
I was having some chats with Kiran Tailor (of Google) and one thing led to another:
We are announcing a webinar/hacking session where I stress-test the latest (Postgres-compatible) Google AlloyDB Omni release using my CatBench (cat benchmarking) vector search stress-test suite!
I created the CatBench Vector Search Playground mainly because I wanted to learn some of the “existing business data + AI” opportunities myself. From the ground up, of course. But I didn’t want to focus on just a single business vertical that few people outside of it understand. So I settled with something that everyone gets - using 25k cat (and dog!) photos from the Kaggle “cats vs. dogs” ML competition dataset. These would be the initial “customers” of my new, modern, AI-enhanced business.
But:
- I also “exploded” these 25k photos to 9 million photos by just rotating every one of them from 1..360 degrees (same pet photo, different angles) - for advanced pet fraud detection use cases that some of us will inevitably have to deal with later on.
- Then, to make all this really useful, I stored and joined all this new cat-customer information (including the Vision Transformer-computed vectors) into an already existing schema of my OLTP transaction processing application. You know, the usual customers > orders > order_items -> products, etc, but now also for cats!
- Thanks to doing that, every time a cat (or a dog) customer visits the main website or any of the nationwide stores, we can use a product recommendation engine at the checkout, simply by looking up what products the other cats - that visually look very similar to our current cat - have purchased! (Yes, by joining a vector search result together with the already existing customers’ purchase history!).
Since AlloyDB is Postgres-compatible, it was easy to port my existing app to run on it. In fact, I plan to also show AlloyDB’s built-in vector query recall monitoring feature, in addition to my home-grown hacks :-)
I have two final messages here:
- Yes, you can augment your existing enterprise applications (where it makes sense) with AI without first having to rearchitect the whole app or build a yet another data lake!
- It is possible to keep the new “AI data” in the same database where all your action is, as long as the database engine and infrastructure can handle the extra data volumes and workload.
And this means lower risk, faster time to market!
The webinar details are here:
Benchmarking High-Performance Vector Search: From Postgres to AlloyDB Omni with CatBench
- Wednesday, July 23th, 2025 at 9:00am EDT
- https://rsvp.withgoogle.com/events/benchmarking-vector-search-with-catbench
See you soon!
P.S. This is just another hacking session like many of my previous ones (not a contractual paid engagement). I’m just trying to make Cat Benchmarking a thing!