For this reason, there are not that many people building concrete Agentic workflows—and for good reason. You need to understand how to set up the agents for your use case and a lot of quality data on the subject.
What we are going to do here, though, is a researcher who can do research in tech up to six months back in time based on a question we give using a custom data pipeline.
This will involve a main agent delegating work to smaller agents that have access to data—or tools, as they are called—that the question requires an answer to.
The workflow will have many steps, involving several agents with access to various endpoints.
You can see the finished demo of what we’ll build using Flowise below.
The finished agent will be able to answer questions about trends in tech, so you should have the capability to talk to it like: 'What are the trending companies in tech for the last month and what is being said?', 'Which platforms have been trending over the last quarter?' or 'what are people saying about so and so?'
You’ll have to test it out on your own to see what it can do.
The following will be a nice learning exercise if you want to try and build these agents. Most of it won't involve much coding using a custom data source. Of course, you can expand on that, working towards improving the agents, adding more tools like search, and hooking it up with other data. I'll go through an intro on agents, costs, building blocks, and data pipelines, but if you're keen just to build, then you can scroll past the intro
I'm going to keep this post rather nontechnical, focusing more on working with the agents through Flowise. You should be able to follow and build the agent even if you're not very technical. What's key here is the focus on how building workflows with Natural Language will differ from doing so programmatically.
You will have to download a GitHub Repo, called Flowise, on which you can run the workflow that we are going to build.
Summed up, it's going to take you about 5 minutes to get up and running, and for this, we are working with the models from OpenAI, meaning you need an API key and at least 0.5 tokens. You may decide on another model.
Everything, save for the tokens, is free because Flowise is open source, too.
System 1 is intuitive, fast, and emotional; System 2 is more deliberate and considered, slower and logical