Name: aiXplain
Uploaded: 2025-01-20T07:50:13.473Z
Duration: 2 min 33 s
Description: A comprehensive AI platform offering tools for building, deploying, and managing AI agents and solutions.

aixplain

A comprehensive AI platform offering tools for building, deploying, and managing AI agents and solutions.

aiXplain

Hi this is Asmar from aiXplain and today I'll be demonstrating how you can build an EIG search agent using aiXplain design. aiXplain design helps you create scalable and production Judy pipelines and automate complex tasks. In this tutorial we'll be learning how to develop an agentic pipeline which is capable of intelligently handling user queries, conducting web searches and providing detailed and contextual responses with minimal coding. 

The core features is that there's an input node which takes in a query from the user and routing system that helps make a decision of whether to follow the first part the second part. The first part provides a direct response via the answer generator and the second follows a query Optimizer and a Google search result for um ex uh for external contexts. Finally we have the contextual synthesizer which allows us to use all of these intermediate results and the original query.

To then formulate a more research fact result. So let's get started. Okay, so we will start with aiXplain design platform with a blank canvas. To begin, we need an input from the user. For that, we're going to get an input node in um. We're going to call this query, and this is going to help us get information from the user that we're going to use with our agents. 

To this, I'm going to attach a model. Uh, we're going to call this model a routing assistant, and this is going to help us um um route the um query from the user. We're going to use a the llm um text generation functionality, and we're going to populate this with an asset from aiXplain Marketplace. Here, I'm using llama 3.1, and this is the hosted by Gro, which is what we're going to use in our model. 

To this model, I'm going to add a prompt. The prompt we're going to use is this: uh, you an expert routing a use a question to either the generation stage or the web search use the web search for questions that require more context for a better answer or recent events. Otherwise, it can skip and go straight to the generation phase to respond. You do not need to be stinged with the keywords in the questions related to these topics. Give a binary choice: web search or generate based on the question. Return the JSON, the single key choice with no preamble or explanation. 

You can customize these options here, but I'm not going to do that right now and connect query to the routing assistant. Uh, and it's going to route the put as a question input for this node to this. I'm going to add a decision node. This decision node is going to basically help us route the decision. It's going to be equal, and the value to compare over here is going to be the choice web search. So if it's web search, it goes to one of the paths or it goes to the other. So I'm going to link this to this.

This is going to be the comparison value while also providing it with the query which is going to be the pass through. I'm going to start with the simpler workflow which is the path one over here. The path one is going to be just a easy answer generator uh using the same functionality of the previous model which is text generation with large language models and llama 3.1 with Croc. 

This is going to connect to this in case the output is false. It's going to be a Simple Text generation output uh while also uh providing the query and everything else over here for this model stays the same. The output we're going to get is going to be a direct response. So if there's a query that does not require a lot of additional context, it's going to directly generate a response and this will just get connected to this and that's our first part. 

For the second part which is slightly more complex, we're going to have a Query Optimizer as our first agent. This we for the function we're again going to select text generation and Lama Gro um llama 3.1 with Croc. 

And over here we're going to uh we're going to provide an additional prompt. The prompted providing for the query Optimizer is that it is an explored crafting web search queries for research questions. More often than not a user will ask a basic question that we wish to learn more about. However, not it might not be in the best format. 

So it's going to reward this query to be the most effective web search string possible and return the outcome with no Preamble or explanation. While everything else Remains the Same, we're going to provide we're going to link the decision note to this in case it's true. 

And we're going to provide the question. This is then going to get connected to a utility node which is our Google search node. We are going to populate this with the

Google search model link this to this. And once the Google search outcome comes with more context and more search results, we would need to format this to be in a more readable format. For which we're going to use a script node for this. I'm just going to upload a script file. The code for this is pretty straightforward and the output data type is going to be text. The input parameters is what we're going to provide and it's going to be context data and it's going to be text. 

So this is going to get formatted under the script node. I'm going to name this post process context and this is what's going to go into the final node, which is the context synthesizer. This is again going to use a large language based text generation model and it's going to use llama 3.1. And once this gets connected to this, we're also going to provide connect the decision node to this as well in case it's true and it's going to.

Give us oh. Before I do that, I do need to provide the context synthesizer with my prompt. The prompt I'm providing is that it is an AI assistant for research question tasks that synthesizes web search results. 

Strictly use the following pieces of web search context to answer the question. If you don't know the answer, just say that you don't know. Keep the answers concise but provide all of the details that you can in the form of a research report. Only make direct references to match if provided in the context. 

Once I do this, you can then come here and select web context as the information being provided as input. Over here, I'm going to provide a question. If it's true, we finally at the end stage where we provide an output node. The output is going to be a reference, and this is the formatted output that we're going to receive. 

Once you're happy with your pipeline workflow, you can choose to share the AI Agent Store with colleagues.

Run this. I'm just going to call this research agent. I'm going to save this and you can try it out here. The prompt I'm going to provide this is that what is the latest news about open a. And there you have it. You have a response generated by the pipeline. If you would like to integrate this into your workflows, you can go into the integration tab and copy the code here and also create your access key under user profile. Thank you for watching.

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

aiXplain Video