perplexity-ai

An AI-powered search engine providing accurate, real-time answers with source citations.

Perplexity AI

In this video we're going to take a team of research AI agents. The research leader, the project planner, a team of research assistants, and an editor will take a topic and automatically produce a well-researched autocitation article that provides a comprehensive view on that topic. 

Let's go ahead and do a demo. I'm going to start my test flow. I'm going to ask for the research to be done on effective parenting methods. I'm going to choose a professional tone. I'm going to choose a thousand words; that's the length of the report I want and six sections to that report, so I'll submit that. 

So what will happen is the research leader will take this topic and it's going to use this perplexity tool. This perplexity tool provides us with the internet search abilities as well as citations. So it's going to execute this flow here and then it's going to now produce a table of contents. 

So this table of contents will serve as the structure, and then the project planner will take that table of contents and it will then map that back into, again, more research on the perplexity tool. And then after it does this, it's going to then now.

Delegate to my team of research assistants. So we'll let this uh continue and and now I have a team of research assistants. So this team will take uh each of the topics that the project planner has created and we'll research each of these individually again using the perplexity tool. 

So now it's processing this. I'll pause as this will take uh a couple of minutes. So it's now completed all the research assistance work and we've aggregated everything together into one text. The editor is going to take that text and it's going to process that into uh one final article and then we're going to create the title and then post. This will take a minute so I'll pause. 

We'll create the title and then uh post to ghost. So this is the final result. It's a well structured article and when you look at how it's structured here the research leader uh they're responsible for planning and doing the initial research that provides this table of contents. 

So this table of contents is a structure that is provided by that AI agent the research leader and then the project planner is the one that's breaking down all these different uh sections. So each of

These sections has been broken down by the uh project planner and each of the research assistants. They are diving into and building just one part of this article and then the editor uh they're responsible for piecing is together. 

But one of the huge advantages of using perplexity and this citation is that these these articles here are real articles. And if you look at all these um descriptions here, all the research it cites exactly where it got this particular information. So this gives you a lot of credibility, a lot of uh confidence that uh the um the AI is not hallucinating, right? 

And then at the end uh there is also a sources as well that uh gives the full list of the citations used. So let's uh do a node by node walk through of how this is built. 

So the first is a form trigger. So this is what specifies the parameters of the research. So I have the form title uh as a topic researcher and then uh what the topic is that we are interested in researching, followed by the tone of the written report as well as then the number of words that we want to produce. 

So this gives us all the parameters uh that can we can customize the.

Report next we start with the uh research leader. So let's go into here. So the research leader uh is the a agent that takes on this Persona and it's quite detailed in terms of the steps that we wanted to take. We wanted to analyze the topic thoroughly and ask it to conduct web research. So this is where the perplexity uh web search and web research uh will come in. It'll automatically call this tool and then we'll give us some structure in terms of uh what information to to gather and then uh what what to create a comprehensive table of contents. 

Right so the table of contents is actually the structure that then uh gets um delegated to the different research assistants by the project planner. Right so then we give it some examples of a project our proposed table of contents and then uh we uh provide more details. Now what's important here is that we're using in here um the GPT 4 mini model. So this is a very cost effective model and for this uh node uh it's enough power for that to do a good job. 

Now the next uh node we're going to talk about is the project planner. So it takes the table of contents that we got from the research leader.

And it's going to uh break down uh that table of contents. So there's some detailed instructions on here on what it's doing. But essentially it's taking table contents and then uh creating a prompt for each of the uh research assistants to uh start the research. Right, so this is effectively uh a delegation of work. 

Now the next note, this looks like a simple note, but it is actually you can think of this as a for Loop. So there's actually it spawns off multiple items. You see here six items, so this uh goes to like six different research assistants even though it looks like one. But there's actually six here. 

So now let's go into the research assistants. So in the research assistant, uh it's actually very simple. Right, it it's just like write a chapter uh for this article as the uh as instruction. But let's look at the Persona. So the Persona is you're a dedicated research assistant uh AI assistant, uh right working as part of a research team under the guidance of the research leader and the project planner. 

And their task is to write a chapter in this overall uh topic. So it goes through uh all the detailed instructions.

What tool to use all right and for for this particular research assistant we're using again the uh gbd4 mini again it's a a good model that uh is actually good enough for this uh case. It's uh doesn't need to be uh that strong in terms of logic. So we're going to use a relatively inexpensive model. 

So after we go through here it merges uh all six uh research uh uh outputs into one and then we do some processing here. Now the editor is extremely interesting note. So let's go ahead and open this up. 

So the editor uh it's an expert editor uh specializing in refining and polishing content right and then it goes through a lot of details. Now what's key here is uh the instructions for th citation. This is um quite quite useful and quite important. 

So what we're telling it is how to deal with citations across all the different research assistant outputs right. So there's a lot of information here on how to create a sources section at the very end of the article for for each of this entry. How do we do that? How do we make it SEO optimize? What the output format should be and and the like? 

So there's a lot of uh details in the prompt and and and.

As such what we're doing here is using a stronger model right. So this is the GPT 40. So this uh 40 is stronger than many uh but uh more expensive right. So we're using this um selectively on uh the the nodes that that needed. 

And then of course we create the title right very simple. We take the output and then create a title from that and then we publish to ghost. Now we've got this flow here. Now let's look at the this tool the perplexity tool. 

So this perplexity tool is from uh perplexity doai. So perplexity doai uh provides a uh online model uh that I'm accessing it actually through open router. So open router uh just makes it easy to access all kinds of models. So I'm just access same perplexity through that and uh I'm I'm using the Oni model which is uh the one that has the the internet search abilities and then citation abilities. 

And what I'm doing here is uh after I call the perplexity I take the output of uh the internet search and then there's also excitations field. So this is very important. So the citations I get back from perplexity and the citations field uh is what's what I use then to uh input back into uh my the.

Different agents that use it. So what this will do is in the case of the research leader, it'll have uh citations in here and research assistants also will have citations. 

And if you look at uh the detailed information for the research assistant, there's a whole uh set of uh uh prompt engineering here that talks about how to deal with citations, how we want to format that. 

So there's a lot of detail in here that uh uses the citation. But overall uh this is a a very powerful workflow and it's it's not too complex just breaking down uh the work uh but very very effective.

Note, that there may be slight differences in text and video. Watch video for the original content.

Perplexity AI Video