Hello, everyone!
I've been meaning to explore Google's AI offerings for quite some time and have been playing around with the various models in AI Studio quite extensively. Lately I've been very impressed by the capabilities of 2.0 Pro for handling context at length.
I have a few assistants projects that I've deployed in OpenAI to date, but to support scalability and to unify the different tools in my AI "toolbox", It makes abundant sense to me to begin the process of consolidating everything on one ecosystem and as a very long time Google Workspace user, it seems like a perfect fit.
I have a couple of projects that would require pairing an agent with a small store of contextual data. At the moment, the size of the contextual data is tiny, it's essentially a GitHub repository with about 20 markdown files, all containing various facets of the project. It's dynamic in the sense that for the agent to operate optimally, I would need to periodically update the context store, but it will never be that large.
I assume that creating something like this in Vertex would be fairly simple, but unless I misunderstood it, and they may well have, the platform is geared towards actually deploying the models on managed hardware.
For the kind of use case I'm envisioning, deploying the models themselves makes no sense. Rather, it's a simple agent-style flow. Ideally, I'd like to have a number of different context data stores, perhaps as many as 10, to segment the data according to the use envisioned for the model. I'd need some easy way to periodically update context and have it added, i.e. something like a pipeline. And at a more sophisticated level, the ability to edit or prune context data storage would also be useful. But for the basic implementation, just having access to the vector stores would be enough.
For the same reason that I like the idea of using Google AI in the first place, I'd also like to have the chatbot interface hosted and available through a Google service. The ability to store conversations in a history would be useful. And as I'm already using Workspace the ability to use that as a context store would be great if structuring it as a formal vector database introduces too much needless complication.
Basically, I love the kind of features that are available in AI Studio, and I'm trying to think of a way to actually bring them quickly into day-to-day use with a little bit of customization that I require.
Many thanks for any recommendations!
LinkedIn
Twitter
