dglazkov
commented
3 months ago Here's an example, generated by https://breadboard-ai.web.app/?board=https://breadboard.live/boards/@dimitri/board-describer.bgl.json
Proposal for the Librarian Board
This board, titled "The Librarian," aims to help users find relevant books based on their interests. It utilizes a multi-step process involving an interview, book searches via the Google Books API, and finally, a summarized list of recommendations.
Here's a breakdown of the components and their roles:
1. Input Node (input):
- This node gathers the initial context from the user.
- It asks the question: "What book are you looking for?"
2. Looper Node (looper-b381bf13 - Interview Planner):
- This node orchestrates the interview process.
- Its task is defined as:
"Based on the initial topic, come up with the themes for 3-5 question interview to collect just enough information to look for an interesting book in the library. These themes will be presented as multiple choice questions with the last choice being an open-ended response."
3. Specialist Node (specialist-2551887d - Interviewer):
- This node acts as the interviewer, posing questions to the user based on the themes provided by the looper.
- Its persona is:
"You are an expert researcher, whose job it is to interview the user to collect information about the kind of book they want. Based on the theme provided and incorporating the history of the interview so far, offer a question that allows the user to easily pick or quickly type an answer."
4. Human Node (human-8b77a262 - Asking User):
- This node allows the user to respond to the interviewer's questions.
- It's configured with the title: "User" and description: "Answer librarian's question".
5. Specialist Node (specialist-27a12855 - Researcher):
- This node leverages the Google Books API to search for books based on the collected interview information.
- Its persona is:
"You are a librarian. Based on the information collected from the interview, formulate the queries (more than one if necessary) and call (in parallel) Google Books API to search for the right book for the user."
- It has the task:
"Make up to three parallel tool calls to Google Books API using different unique phrases to get the best results."
- It utilizes the tool: "https://breadboard.live/boards/@dimitri/tool-google-books.bgl.json"
6. Specialist Node (specialist-8928cd1a - Summarizer):
- This node summarizes the search results and creates a list of recommendations.
- Its persona is:
"You are an expert librarian. You take the raw book search results from multiple queries and the interview information that was collected, and come up with a list of book recommendations."
- It has the task:
"Come up with a list of 5-7 recommendations. Reply in markdown, making book titles hyperlinks to the book's canonicalVolumeLink and use smallThumbnail to display the book's thumbnail. "
7. Output Node (output):
- This node presents the final list of book recommendations to the user.
Workflow:
The user provides initial context about the desired book. The looper node then generates interview themes, and the interviewer node poses questions based on these themes. The user responds, and this information, along with the initial context, is used by the researcher node to query the Google Books API. Finally, the summarizer node compiles the results into a user-friendly list of recommendations, which are displayed via the output node.
This board effectively combines human input with AI-powered tools and specialists to provide a personalized book recommendation experience. The use of the looper node allows for a dynamic and iterative interview process, while the integration of the Google Books API ensures a comprehensive search for relevant titles.
Now that we have plenty of boards as examples, let's build a board that generates boards.