This beginner-level project showcases a multi-agent LLM orchestration system that coordinates specialized agentic AI models to provide personalized book recommendations and deliver an inspiring quote for your daily dopamine boost.
Note: This project is for demonstration purposes only.
A single AI chatbot often struggles with complicated tasks, which frequently leads to errors and hallucinations when the model becomes overwhelmed.
So, let's fix this! My project uses a team of AI agents that work together just like a real-life human team. Instead of asking one AI chatbot to do everything, we split the work among a set of specialized agents. Each agent has its own specific role, rules, and tools, which gives us higher quality results and reduces the chance of errors.
What makes a single AI agent different?
An AI agent can function autonomously with only a set of instructions and the tools you provide. By assigning it a goal, it can complete difficult tasks and eliminate the need for repetitive work.
Agents:
Coordinator - Acts as the manager of the entire system. Instead of executing specific tasks like surfing the web or generating content, it orchestrates the other agents to ensure the user's goal.
Research - This agent acts as the book researcher of the group. Its main goal is to find the highest quality book recommendations and hand the results over to the coordinator.
Quote - This agent acts as the quote specialist. Its sole purpose is to find high quality quotes from any of your favorite writers. You can also just ask for the 'quote of the day' to get a random one as well!
The agents are completely capable of functioning autonomously, but in this specific demonstration, they only work when called upon. The research and quote agents can't communicate with each other directly, so all messaging is handled strictly through the coordinator agent.
Topology Diagram:
The magic happens within the Hub! Requests are routed using a transitional graph, and the agents only respond when called upon so mistakes don't happen!
Results:
