0:00 Welcome to session two of the Funder Research and AI Agent Pilot. I am Chris from Ground Force Digital. We are still in the learn phase of our four-step sprint.
0:14 In our last session, we explored the raw data. Today, we are going to explore the tools we will use to put that data to work.
0:22 The agent platforms. Our agenda for today is straightforward. We'll start with a quick recap of the data transformation tools we discussed in session one.
0:33 Then, we will dive into a core decision of this pilot. Which low or no-code platforms to use. And we'll review a few specific platform options and select two to proceed with for the design phase of the project.
0:48 First, a quick look back at session one. You should have received the session kit, which includes the presentation slides and the Google Colab walkthrough.
0:59 As we discussed, the Colab document is a step-by-step guide to transforming the open government data into a clean format. It bridges the gap between raw spreadsheets and a dataset our AI agent can then actually understand.
1:14 If you haven't explored it yet, I encourage you to take a look at the background materials in the kit. Now, let's talk about building the agent.
1:24 For this pilot, we are going to select a visual no-code platform, but I would like to spend a moment comparing them to developer platforms, also commonly called developer kits for agent building.
1:36 We're going to use visual platforms, which are designed for rapid prototyping. They allow non-technical teams to build chains of thought by dragging and dropping blocks, much like building a flowchart.
1:51 This linear workflow makes the logic easy to see and understand. In contrast, developer platforms offer granular flowcharts, and are powerful for complex math or data analysis.
2:06 However, they have a high barrier to entry. They require knowledge of Python, APIs, and environment management. For many non-profits, this creates a dependency on external technical providers.
2:20 While visual platforms can sometimes struggle with very complex, Thanks. unstructured data or debugging, they allow us to go from an idea to a working prototype in hours, not weeks.
2:33 For a pilot like this, their functionality, speed, and accessibility is exactly what we need. So, which platforms are we looking at?
2:43 Let's consider three options. OpenAI's Agent Builder, Flowise AI, and Google's Vertex AI. OpenAI's AgentBuilder is our low-barrier option.
2:57 It is highly accessible and allows you to build an agent using natural language instructions. Importantly, it's a builder offered by a well-known and widely adopted market leader.
3:10 Flowise AI is our middle ground. It is a dedicated visual builder that connects different large language models to your data.
3:19 It requires a bit more understanding of logic flows, inputs and outputs, but offers more customization than OpenAI without requiring you to write code.
3:30 Finally, we looked at Google Vertex AI. While powerful, it is an enterprise-grade tool. For the scope of this specific pilot, we found the administrative overhead, managing cloud billing, projects, and permissions, to be too high.
3:47 Therefore, we will focus our testing on OpenAI and FlowWise. As we wrap up the learn phase and prepare to start establishing goals for the agent build and sketching solutions, we can think about how these can connect to other data sources beyond the charity's open data set, such as external web content
4:08 and our internal organization data stored in systems like our CRM, email, and shared drives. In our next session, we will move into the design phase.
4:20 We will define more specific goals for the pilot and compare notes on the solution will be. I look forward to seeing you next week.