RAG Agents are designed to retrieve information from a multitude of sources, including databases, documents, and APIs, and then use this information to generate an answer.
This pattern will focus on Retrieval from unstructured data, which comes with a lot of challenges including the need to understand the context to deliver on accuracy, performance and cost of operation.
We put RAG in perspective within the context of an Agentic AI System, to showcase how complex problems can be solved by a RAG enabled agent.
🚀 Retrieval Augmented Generation (RAG) Accelerator
This pattern covers the design and implementation of agents that can generate code snippets, functions, or even entire programs based on a set of requirements. This pattern is particularly useful for developers who want to increase their productivity while respecting existing code standards, libraries, reference architectures, and best practices.
The applicability of a Code Generation Agent is vast, and we will cover the most common scenarios, namely:
- coding assistance
- code translation
- automated testing
🚀 Code Generation Agent Accelerator
This pattern explores best practices for building Multi-Agent Systems (MAS) that consist of multiple agents. Each agent focuses on a specific functional area, and they are coordinated to combine their capabilities, offering a unified, enhanced super-agent experience to the end user. The primary goal is to scale the agent system across multiple functional domains. The pattern will address several key aspects of MAS, including:
- Decomposition of the overall system's functional scope into more manageable segments.
- Designing mechanisms for routing and forwarding.
- Management of shared memory and state transitions across agents.
- Handling the transition of communication flow from one agent to another.
They also enable pausing tasks in one domain to switch to another domain and then resume the task in the original domain if needed to complete a complex workflow, or if directed by the user.
The pattern will cover some key-aspects of this pattern, including:
- domain scoping & detection, domain switching, request delegation
- long-running contextual memory: memorizing context accross domains to be able to resume task started in domain A when user switches to domain B and then back to domain A