Starting this spring I committed to get on top of LLM Agents and adjacent technologies.

With a certain degree of experience in learning new things, I started with absorbing all of the structured knowledge I could get my hands on: blogs, courses, open source projects and many more.

I got previous experience with LLMs, transformers and NLP in general. So this time I focused on one thing only: how to empower LLMs to do stuff for me.

Starting small and easy, I want to share courses that helped me shape decent foundation of understanding and putting Agents to Practice.

With a bit of excitement and my tendency to lean towards open source, I chose probably the biggest proponent of open source AI, Hugging Face, to help me dip my feet in the water of LLM Agents with their course https://huggingface.co/learn/agents-course/unit0/introduction.

I loved pretty much everything about this course. Especially the challenging and practical final assignment focused on building general-purpose agent and testing it on GAIA dataset.

This course is built on practical examples of building on top of Langchain, LangGraph, Smolagent (Agent framework from HuggingFace that allows you building Code Agents in less than 5 lines of code) and LlamaIndex.

Next on, thanks to amazing OLX Group support, I got my hands on 2 courses on AI Agents  from Udacity. Courses stood out in their emphasis on strong foundations and supporting technologies for AI Agents. ((https://www.udacity.com/course/generative-AI-with-AWS–cd13232 and https://www.udacity.com/course/intro-to-building-generative-AI-solutions–cd13267)

Udacity lecturers outlined the importance of ethical and responsible AI use. Building on top of that, I learned much more about vector databases, multimodal LLMs and practiced using AWS tools to fine-tune foundational LLMs for domain-specific requests. As the result, I got a few quite competent AI agents in Finance and Real Estate domains.

To polish it off, I got back to HuggingFace’s course on MCPs (https://huggingface.co/learn/mcp-course/unit0/introduction) to bring structure in integrating agents with tools. Because, it got messy at certain point when had to connect APIs with different schemas from different providers and had to get LLM to understand and use them well. This course clearly explains what all of this hype is about and teaches using MCP for tools integration by creating your own MCP servers and clients.

What’s next?

Put skills to practice on small and low-risk project. Next… scale project and maybe teach someone else to solidify acquired skills 😀

I have something in fitting mind already and probably won’t return to theoretical courses for a while. However, I am always happy to read a shorter article from time to time.

Do you have your favourite Agents related article? Share in the comments and help me learn.