Izvajalec / Course provider: Faculty of information studies in Novo mesto (FIŠ)
Predavatelji / Instructors: Biljana Mileva Boshkoska (FIŠ), Robi Pritržnik (FIŠ), Pavle Boškoski (FIŠ)
This intensive 8-hour course bridges the gap between static Large Language Models and dynamic organizational knowledge. Participants will learn how to build Retrieval-Augmented Generation (RAG) systems on high-performance infrastructure. The content covers the entire architecture, from converting private documents into vector embeddings to utilizing frameworks like LangChain. Upon completion, participants will be equipped to deploy a secure system that provides accurate answers without the risk of model hallucinations.
Learning objectives: Participants will gain a thorough understanding of the RAG architecture and its advantages over model fine-tuning. Through hands-on work, they will master the data preparation process, including text chunking strategies for optimal information retrieval. They will learn how vector embeddings function and how to select appropriate models for the semantic processing of documents. Participants will gain concrete experience in setting up vector databases for high-speed content searching and learn to automate workflows using the LangChain framework, including generating responses with precise source citations.
Course content: The course begins with an introduction to RAG infrastructure and the role of context in enhancing the relevance of AI responses. This is followed by a module on data pipelines, covering document cleaning, chunking strategies, and metadata tagging. The core part of the course focuses on working with vector databases (e.g., Pinecone, Chroma) for semantic search. In the practical LangChain segment, participants build chains, manage conversational memory, and optimize search results. We conclude with a chapter on security and evaluation, testing the system's robustness on real-world business cases.
Learning outcomes: After the course, participants will be able to independently establish a functioning RAG system that processes private company documents in real time. They will acquire the knowledge to choose the optimal technology stack based on the technical and security requirements of their organization. Effective use of the Python language will enable seamless integration of models with their own knowledge bases. With the knowledge gained, participants will reduce operational risks by ensuring factual accuracy and information traceability, becoming qualified to lead digital transformation projects.
- 9:00 – 10:30 | From Documents to Vectors: The RAG Foundation Learn how to bridge the gap between static LLMs and dynamic data. We will cover the conversion of private organizational documents into vector embeddings—the mathematical language of AI knowledge.
- 11:00 – 12:30 | Managing Knowledge: Vector Databases on HPC Explore the management of high-speed vector databases. Learn how to store, index, and query massive datasets efficiently using the high-performance infrastructure of HPC Trdina.
SESSION 2: Orchestration and Deployment
When: Friday, November 20, 2026 | 9:00 AM – 1:00 PM (including break)
Where: UNM FIŠ, Ljubljanska cesta 31a, 8000 Novo mesto
- 9:00 – 10:30 | LangChain Orchestration and Retrieval Pipelines Master the orchestration of retrieval pipelines using Python frameworks like LangChain. Learn to connect the user’s query with the right data to provide context-aware responses.
- 11:00 – 12:30 | Launching the Pipeline: Hallucination-Free AI Practical deployment of a functional RAG system. Learn to fine-tune the pipeline to eliminate AI hallucinations and ensure accurate, data-backed responses based on specific datasets.
Why attend?
Hands-on experience on a real system: Design and launch a functional RAG pipeline directly on HPC Trdina.
Free knowledge: Training is part of the SLAIF initiative to enhance digital competencies.