The microcredential in Transformers, LLMs, RAG and Agents: From Theory to Production offers practical training in modern GenAI technologies, covering the entire spectrum from transformer architectures to autonomous agents. Participants will learn to work with large language models (LLMs), build sophisticated RAG systems, implement the Model Context Protocol, and create ReAct agents capable of reasoning and acting in complex situations. The course emphasizes practical implementation alongside theoretical understanding, preparing participants to deploy production-ready AI solutions.
A key focus of the course is on designing RAG systems with vector databases, fine-tuning domain-specific language models with Hugging Face, and building MCP servers to extend agent capabilities. Participants will also learn to evaluate deployment trade-offs between on-premises and cloud APIs while optimizing, debugging, and maintaining LLM applications for seamless software integration.
In this intensive program, participants will gain a strong base in state-of-the-art GenAI engineering, preparing them to master complex workflows and enterprise-grade systems. This program is designed to strengthen your professional profile within the fields of artificial intelligence, machine learning, and advanced software engineering.
*IMPORTANT NOTICE: Please note that enrolling in this course implies a firm commitment to complete it in its entirety.
Asynchronous online format.
PhD in Computer Science and Associate Professor at FIB-UPC. He is a member of the EduSTEAM research group on engineering and science education, and author of publications in areas as diverse as educational technology, artificial intelligence, and ethics in engineering, as well as e-learning and software architecture. He is also known for his work as a tech podcaster and creator of popular digital platforms such as Mossegalapoma.
Learning Methodology
The microcredential is taught in an asynchronous online format through a virtual environment that combines interactive content, authentic materials, and applied activities.
Learning is based on practical case studies and real-world situations that connect theoretical concepts with professional decision-making.
Instructors provide ongoing guidance and personalized feedback on the proposed tasks and assignments.
Training Content
- Introduction to Natural Language Processing (NLP) and transformer architecture.
- Use of the Hugging Face library and fine-tuning of pre-trained models.
- Understanding LLM capabilities, emergent properties, and prompt engineering.
- Model deployment via APIs and locally using Ollama and llama.cpp.
- Overview of multimodal language models.
- Principles of semantic search, vector embeddings, and vector databases (such as Pinecone or Chroma).
- Design of RAG (Retrieval-Augmented Generation) architectures and assistant development.
- Optimization strategies and comparison between RAG frameworks and custom developments.
- Fundamentals of the Model Context Protocol (MCP) and integration with development tools (VS Code, Cursor, Claude Desktop).
- Creation, debugging, security, and deployment of custom MCP servers.
- Development of autonomous agents based on the ReAct (Reasoning and Acting) paradigm.
- Agent orchestration, Docker containers, and introduction to the A2A (Agent-to-Agent) protocol.
- Solution integration with Google and OpenAI APIs.
Prerequisites:
Students are expected to have basic knowledge of:
- Basic to intermediate programming skills in Python, HTML, and JavaScript
- Experience with Git and command-line tools
- Basic knowledge of APIs, web services, and HTTP protocol
Certificate/badge details
Microcredential. Europass digital credential in Transformers, LLMs, RAG and Agents: From Theory to Production issued by the Universitat Politècnica de Catalunya.
