Agentplex Weekly - Issue #4
Building Agents with DSPy. Multi Agents with crewAI. Agentic RAG with LlamaIndex. Agentic Design Patterns with AutoGen. Symbolic Chain-of-Thought. CodeActAgent. PromptWizard.
Why building AI Agents with DSPy? DSPy is not an agent framework, but it can enable you to build agent pipelines programmatically without dealing with prompts and to tune these pipelines in a LLM-agnostic way. DSPy separates the flow of your program (modules) from the parameters (LM prompts & weights) of each step. In addition, DSPy optimisers (LLM-driven algos) can tune the prompts and/or the weights of your LM calls based on a metric. Let me share 3 cool blogposts on DSPy and building agents:
Tutorial: A first attempt at DSPy agents from scratch. This is a great walkthrough on building an agent app using DSPy. By the end of this post, you'll learn why DSPy is great for creating a functional agent system that includes: Plans, Workers, and Tools. Blogpost: A first attempt at DSPy Agents from scratch.
Building an AI agent with DSPy. Do you hate prompt engineering? What if, though, you can program on top of LLMs using a high-level programming framework, and let the framework write and tune prompts for you? In this post, Lak writes about how to do zero shot prompting, agents retrieval, optimisation, and much more. Blogpost: Building an AI assistant with DSPy.
The ReAct Pattern and Multi-Agent DSPy Programs. This is a great notebook showing one complex strategy that DSPy makes near-trivial to achieve: Automatically bootstrap five different highly-effective prompts for ReAct, then optimise an aggregator that combines their powers. Notebook here: Multi-Agent DSPy ProgramsBootstrapping & Aggregating Multiple ReAct Agents.
Three free courses
Multi AI Agent Systems with crewAI. Learn more about the key components of multi-agent systems like: role-playing, memory, goals & planning, guardrails, and cooperation. You’ll also learn how to design & build multi-agent systems that execute common business processes. Free course: Multi AI Agent Systems with crewAI.
Building Agentic RAG with LlamaIndex. Learn how to build a research agent with agentic RAG and routing, that is capable of intelligently navigating, summarising, and comparing information across multiple research papers from arXiv. Free course: Building Agentic RAG with LlamaIndex
AI Agentic Design Patterns with AutoGen. In this course you’ll learn how to build and customise multi-agent systems, enabling agents to take on different roles and collaborate to accomplish complex tasks using AutoGen, the MIT-licensed framework from MSR that enables development of LLM applications using multi-agents. Free course: AI Agentic Design Patterns with AutoGen.
Three interesting research papers
Symbolic Chain-of-Thought (SymbCoT). This paper argues that integrating symbolic expressions and logic rules with CoT prompting improves LLMs reasoning and explainability - Checkout the paper & repo here: Faithful Logical Reasoning via Symbolic Chain-of-Thought.
CodeActAgent. Introducing a chatty coding agent fine-tuned from Llama2 and Mistral, that is integrated with the Python interpreter. No more JSON! It executes multi-turn code tasks dynamically. Checkout the repo, CodeActInstruct dataset, demo and paper here: Executable Code Actions Elicit Better LLM Agents.
PromptWizard by MSR. An agent that optimises both prompt instructions and in-context examples, maximising model performance. It iteratively refines prompts by mutating instructions and incorporating negative examples to deepen
understanding and ensure diversity. Paper: PromptWizard: Task-Aware Agent-driven Prompt Optimisation Framework.
Update on the Agentplex AI Agents Global Challenge
Wow! We keep receiving some amazing applications. Remember, you can still keep refining your applications up to the final submission deadline, which we plan to extend a little bit beyond the original deadline of 1 September.
Based on your feedback, we will soon be announcing important further details, including a better allocation of the $1M prize pool, more benefits for the participants, and some guidelines and ideas. To receive updates and further announcements, keep reading this newsletter, join our Discord channel, follow us on Agentplex X channel, or simply drop us an email.
Upcoming AI Agents hackathons and meetups
Starting in London, and later expanding to other locations, will be announcing our first AI Agents Meetup and AI Agents hackathon soon. If you are interested in giving a talk or do a demo at a meetup or suggest ides for a hackathon, please contact Carlos me here.
If you are interested in a summer internship, please submit an email here to apply.
Thank you for reading Agentplex newsletter. Have a great day.