Stop Buying “AI Platforms” — Build the Stack First (Stack Builder Walkthrough)
Transcript
is you could use a tool to basically input your AI infrastructure requirements and it spits out a report with timelines and pricing risk etc that you could just hand over to your team and say decompose this. It wouldn't even be better if it was based off of a wellestablished AI infrastructure infrastructure or AI infrastructure model. You know what? Such a tool does exist. It is the CTO Advisor 4 plus one stack builder and it is based off of my IP around the 4 +1 AI infrastructure model that I released a little bit over a month ago of the recording of this video.
com. Either click on stack bidder or hash stackbuilder in your URL. We're going to begin building the layers. It is based off of nine simple questions I guess from a wording perspective, but these are again decision points that you have to make or that you have to feed it based on your existing infrastructure. We're going to go with one of the simplest routes and say that we're all in cloud. We're going to choose Google Cloud as our primary cloud. We are going to dictate what we're going to use our AI infrastructure for.
We're going to use it for retrieval applications and inferencing at scale. We're not going to do much training or fine-tuning. We're going to say that we have a new small team. We don't have a Kubernetes cluster. This is again basically uh an existing environment, but we're not going to use any of our legacy infrastructure or resources. You can and you'll get a much more robust report as a result. Budget. Let's start with a modest budget for AI 100,000 to 500,000.
We are going to do this in the next the next quarter. Again, this is to assess risk, some historical data collection to help me understand what people are doing out there. Our number one concern because we are using cloud is developer velocity. What are our considerations from a operations perspective and you know high data egress costs are always and of course we really want our developers not waiting on infrastructure. So again, developer velocity and you know what for the my security and compliance related folks out there, you know what?
Let's let's go with agents can access government data safely. All right. So this gets us to what do we need to build and what are the critical layers. always layer 2C is a must-have requirement in my framework generally speaking and any additional required layers. And now we're going to do a vendor search for what's out there. This is a real time search. This isn't based off of a fixed model. in the background. 5 Pro to do a active search for each layer that is going to come back with set of solutions and a highlevel cost model for each solution we select based on what public information the AI will retrieve.
So with that said, it takes a relative long time to when you're using AI with let's say a chat GTP or perplexity. It takes about 2 minutes or a minute and a half as you can see. So obviously we're going to fulfill a lot of our requirement with Google Cloud Vertex. You see the interface update automatically with with what's missing. So we've covered our 2C, our 1A and our 2B. We now need to cover our 2A infrastructure orchestration. And here we can go with some type of best of breed solution.
Let's go run AI on GKE. In addition to that, let's say that there's a solution that you have on premises already that is not shown. We can ask the AI a question. Virtual CTO advisor. We can say we are a NetApp shop and need NetApp for our storage leader. Hit send. There will be an option to regenerate the selection based on our chat session with virtual CTO advisor. So yes, this is integrated with virtual CTO advisor. Once we have all of our vendors selected, we'll go to continue with selected vendors.
It'll validate that we've had all of our layers covered. And then we'll generate the road map. Generating the road map takes roughly about 2 minutes. Again, this is some pretty heavy lifting we're doing in the background. All right. And here's our report. This is the HTML form of it. We have everything you would expect in a detailed report from a virtual CTO or full C time CTO. Estimated cost, timeline, risk, failure modes, uh your 2C reasoning plan, audit.
I'm playing around a little bit with the scoring on this, but again, your 2C audit, your overall architecture, a cost analysis, and you can download this as a PDF. There is no type of information wall. Let's say you have further questions. You can ask virtual CTO advisor additional questions around the complexity or clarifications within this report. I'd love to hear feedback about this tool. What do you think about it? Is it useful? com.
