Trainings

RAG and GraphRAG Training – Combining Semantic Search with Linked Knowledge

Upon request

1–2 days

Online, in Munich, or in-house

Upon request

German or English

Digital Certificate of Participation

Combine RAG and GraphRAG to uncover semantic relationships and provide generative AI with relevant context. Generative AI does not automatically have access to the internal information, current documents, or domain-specific knowledge required to produce reliable results. Retrieval-Augmented Generation (RAG) retrieves relevant information from connected knowledge sources and provides it to the language model as context. Vector-based retrieval identifies text passages with similar content. GraphRAG supplements this approach with explicitly modeled relationships between entities and concepts—ensuring that not only individual text passages but also cross-document relationships can be incorporated into the response.
Our RAG and GraphRAG training course combines the development of a complete RAG pipeline with a graph-based extension. You’ll learn about data preparation, embeddings, chunking, semantic search, vector databases, and evaluation, and implement the pipeline in Python notebooks. Building on this foundation, you’ll identify entities and relationships in a knowledge graph and combine them with the relevant original texts to form a unified answer context. In doing so, you’ll explore both open-source LLMs and commercial models and work with your own or provided data sources.

Contact Person

Jonas Kuhlmann
Top Consultant Award
Satisfied customers from SMEs and large corporations

Content and Schedule of the RAG and GraphRAG Training Courses

The training course follows a path from the fundamentals through technical implementation to evaluation. You will first develop the RAG foundation and then expand upon it with GraphRAG during the second part of the technical implementation. The focus is on the interplay: finding semantically appropriate content, identifying relevant relationships, and combining both into a coherent context for the answer.

  • Introduction to Generative AI and the Role of RAG in Modern AI Architectures
  • Understanding the Interplay Between Knowledge Sources, Semantic Retrieval, Knowledge Graphs, and Text Generation
  • Classifying RAG, vector-based retrieval, and GraphRAG: GraphRAG as an extension of RAG, not as a competing underlying principle
  • Assess when semantic search is sufficient and when cross-document relationships or multi-step queries make a graph-based extension useful
  • Integrate documents, databases, websites, wikis, and APIs as knowledge sources
  • Classifying Different Integration Paths and Data Access Methods
  • Selecting Appropriate Large Language Models for RAG Applications
  • Comparing Open-Source LLMs and Commercial Models
  • Distinguishing Between RAG and Fine-Tuning
  • Analyze typical sources of error, quality issues, and best practices

Result
: You can classify your own use cases and outline a knowledge architecture that combines semantic retrieval with graph-based context as needed.

Part 1 – Semantic Search and the Complete RAG Pipeline

  • Organize and structure documents and knowledge resources
  • Understanding the Basics of Embeddings and Semantic Similarity
  • Develop chunking strategies and evaluate their impact on retrieval quality
  • Visualizing and Interpreting Similarities in the Embedding Space
  • Using Similarity Search, Top-K Retrieval, and Vector Databases
  • Use reranking strategies to improve the relevance of search results
  • Developing prompt designs for context-based and transparent responses
  • Pass retrieval results to the language model in a structured format
  • Implementing a Complete RAG Pipeline in Python Notebooks
  • Work with your own or provided data sources

Part 2 – GraphRAG: Uncovering Relationships and Combining Them with Semantic Search

Semantic similarity answers the question, “What content matches the query?” The knowledge graph adds, “Which entities are related to one another, and how?” A combined retrieval architecture utilizes both perspectives: It identifies semantically relevant entry points and supplements them with linked entities, relationships, and associated original passages. This creates an answer context that brings together content-based proximity and explicit connections.

  • Extract and standardize entities and concepts from texts, and model their relationships as nodes and edges
  • Make the meaning of a relationship explicit, for example, “is part of,” “uses,” or “depends on”; do not equate similarity with a documented relationship
  • Use semantic retrieval as a starting point and expand the context using related entities and relevant source passages
  • Process multi-step queries across multiple relationships (multi-hop) while maintaining the connection to the original sources
  • Using Microsoft GraphRAG as an example, distinguish between local, entity-based search and global search via summaries of thematic graph communities
  • Prioritize, filter, and consolidate text matches, entities, relationships, and, where applicable, community summaries within the available context window
  • Hands-on: Extend the existing RAG pipeline with graph-based retrieval and compare both approaches on the same set of questions
  • Generate responses with references and verify whether the relationships presented are supported by the underlying documents

Illustrative practical example: A question might be: “Which products could be affected by a change to component A?” The semantic search finds relevant change documents. The graph supplements the modeled relationships between the component, assembly, and product; linked original passages provide the supporting evidence. The language model can use this information to generate a well-reasoned answer—provided these relationships are present in the knowledge base and have been correctly recorded.

Result
A functional RAG pipeline with graph-based expansion: semantically relevant content and explicit relationships are linked to form a source-based response context.

  • Systematically Evaluate Retrieval and Response Quality
  • Develop appropriate test data sets and evaluation metrics
  • Define quality metrics for retrieval, source attribution, and generated responses
  • Implement Automated and LLM-Based Evaluation
  • Targeted Optimization of Chunking, Embeddings, Retrieval, Reranking, and Prompts
  • Compare Parameters and Conduct Experiments in a Reproducible Manner
  • Experiment tracking and versioning, for example, using MLflow
  • Deploy a RAG application as an API service using FastAPI
  • Understanding the Basics of Deployment, Monitoring, and Operations
  • Detecting Data and Concept Drift in RAG Systems
  • Analyze changes in data sets and develop countermeasures
  • Shaping the transition from prototype to a maintainable, production-ready solution
  • Compare vector-based RAG and the combined RAG/GraphRAG pipeline on the same test questions: individual facts, cross-document relationships, and overarching topics
  • Check extracted entities and relationships for completeness, ambiguity, and source references; consider incorrect links as a potential source of error
  • Weigh the additional indexing effort, response latency, and costs against the measured improvement in quality—GraphRAG is not automatically the better choice for every query

Results
: An evaluated RAG/GraphRAG approach featuring a transparent comparison of answer quality and computational cost, as well as prospects for API operation, monitoring, and production use.

What You Can Put into Practice Right Away After the RAG and GraphRAG Training

After completing this training, you will be able to design RAG systems, extend GraphRAG, and evaluate the benefits of this combination based on specific questions. You will be able to distinguish between semantic similarity and explicit relationships and integrate both into a coherent response context.

Who is the RAG and GraphRAG training course intended for?

This training is intended for:

  • Data Scientists and Machine Learning Engineers
  • Software Developers and Data Engineers
  • AI and Innovation Teams
  • Solution and Software Architects
  • Head of Knowledge Management and Digital Products
  • Technical Project Managers and Product Owners
  • Companies that want to combine generative AI with their own data

This continuing education course is suitable for participants with a basic understanding of technology and some experience with Python or generative AI.

Prerequisites for the RAG and GraphRAG Training

Basic knowledge of Python, software development, or data science is recommended. Some experience with large language models or generative AI is helpful, but not a requirement.

Custom RAG and GraphRAG Training Sessions and In-House Workshops

About Us as a Training Provider

We are an experienced management consulting firm with expertise in artificial intelligence, data science, software development, and digital transformation. Our RAG training courses combine technological fundamentals with practical implementation and demonstrate how generative AI can be safely integrated with corporate knowledge and your own data.
We help companies develop RAG systems, from the initial concept through prototyping and evaluation to API integration, monitoring, and production-scale deployment.

01

Practical rather than theoretical
Our training courses are based on real-world project experience, not just textbook knowledge.

02

More than 20 years of experience
We understand the requirements and challenges of numerous industries and incorporate valuable insights directly into your training.

03

Practical, ready-to-use expertise
Everything we teach, we put into practice ourselves every day with our clients—guaranteed to be practical and goal-oriented.

04

Tailored to Your Business
Content, examples, and tools can be tailored precisely to your processes and challenges.

05

Trainers with a background in project work
You’ll learn from consultants who actively lead and support projects themselves—not just trainers.

06

State-of-the-art methods and technologies
You'll benefit from the latest trends, proven best practices, and practical tools tailored to your situation.

Expertise in Retrieval-Augmented Generation – Your Trainer

Jonas Kuhlmann

Jonas Kuhlmann helps companies develop and implement modern AI systems. His areas of expertise include generative AI, data integration, and the implementation of productive RAG applications. He explains technical concepts in an accessible way and guides participants through a structured process, from use case evaluation to a fully functional pipeline.

Jonas combines in-depth knowledge of LLMs, data preparation, and software development with a keen eye for quality, scalability, and productive use. As a result, you’ll not only learn how to develop a RAG prototype, but also what’s required for stable operation within an organization.

Inhouse Academy: Build expertise. Measure impact. Strengthen the organization.

Upon request, we can develop a comprehensive learning program for you that combines individual training sessions with structured learning paths, internal champions, and digital application of what you’ve learned.

Get a no-obligation
Request an Appointment

TISAX and ISO certification apply only to the Munich location

Your message



    *Pflichtfeld

    Bitte beweise, dass du kein Spambot bist und wähle das Symbol Stern.

    TISAX and ISO certification apply only to the Munich location

    Discover more continuing education courses, seminars, and training programs

    ChatGPT Training: Using AI Professionally in Your Daily Work

    Less effort, more impact: By using ChatGPT strategically, you can turn daily routines into real value creation.

    ChatGPT and Copilot Training: Save Time in Your Daily Routine with AI and Effective Prompt Engineering

    In this training session, learn how to use ChatGPT, Copilot, and targeted prompt engineering to efficiently integrate artificial intelligence into your daily work—for greater innovation, value creation, and productivity.

    Product Owner Training: Your Guide to Successful Scrum Teams

    In this training course, you’ll learn how to fulfill your role as a Product Owner strategically and effectively—for greater clarity, focus, and sustainable success in day-to-day agile project work.

    Resilience Seminar for Executives

    In this resilience seminar for executives, learn how to use tried-and-true methods and effective routines to strengthen your own resilience and that of your team—for greater confidence, clarity, and stability in challenging leadership situations.

    Release Manager Training

    In this training, you’ll learn how to establish and optimize structured and agile release management processes to ensure faster and more reliable service delivery in your organization.

    Requirements Engineering Training: The Key to Successful Projects

    In this training course, you’ll learn how to create clearly defined requirements and, as a result, make your projects more efficient, more customer-focused, and less risky—for demonstrable success right from the start.

    Resilience Training for Executives, Teams, and Employees: Targeted Strengthening of Resilience

    In this training course, you’ll learn how to use tried-and-true methods to strengthen your personal and collective resilience, effectively reduce stress, and confidently navigate changes in your day-to-day work.

    Training: How to Attract, Lead, and Retain Generation Z

    Find out here how to apply modern leadership approaches to successfully inspire, attract, and retain young Generation Z talent in your company.

    Scrum Basics Training – Agile Fundamentals for Sustainable Business Success

    In this course, you’ll learn the basics of agile work using Scrum. You’ll gain practical insights and learn how to effectively use modern project management methods and apply them directly.

    FAQ – Frequently Asked Questions About RAG and GraphRAG Training

    Retrieval-Augmented Generation is an approach in which a language model retrieves relevant information from external knowledge sources and uses it to generate responses. This allows the AI to access up-to-date, company-specific, or domain-specific content. As a result, the responses are more closely linked to the provided sources than when using the language model on its own.

    GraphRAG is a graph-based approach within Retrieval-Augmented Generation. In addition to text passages, entities and their relationships are identified to provide context for the response. To do this, Microsoft GraphRAG generates a knowledge graph from texts and utilizes, among other things, summaries from thematic communities. Local search links graph information to relevant original passages. In this way, GraphRAG complements RAG by providing structured access to interconnected knowledge; it does not replace the documents or their source references.

    Semantic search identifies content or entities whose meaning matches the query—even when the phrasing differs. These results can serve as entry points into the knowledge graph. From there, relevant relationships and associated text passages are identified and prioritized for the language model. In this way, the response combines thematic relevance with documented connections, rather than relying solely on similar text excerpts.

    This combination is particularly worth exploring when answers need to combine information from multiple documents, trace relationships across multiple steps, or explore overarching themes. For simple questions that can be answered locally, vector-based RAG may be sufficient. A knowledge graph requires additional effort to build and maintain; missing or incorrectly extracted relationships can compromise answers. Therefore, during training, evaluate the actual added value using appropriate test questions rather than automatically assuming higher quality.

    A basic knowledge of Python or software development is recommended for the technical implementation. In addition to the concepts, the training covers a complete pipeline in Python notebooks. Participants without programming experience will be able to understand the architecture and how it works, but will have limited ability to implement the practical aspects on their own.

    Typical sources include documents, PDFs, databases, websites, wikis, and APIs. In the training, you will learn how to prepare and structure content and make it available for retrieval. Your own data sources can be incorporated into in-house training formats.

    RAG supplements a language model at runtime with relevant information from external knowledge sources. Fine-tuning, on the other hand, modifies the model itself through additional training. During the training, you will learn for which requirements RAG, fine-tuning, or a combination of both approaches is appropriate.

    Embeddings convert content into numerical representations that can be used to calculate semantic similarities. Vector databases store these representations and enable fast searches for relevant content. Both components form an important foundation for high-performance retrieval.

    Quality is assessed on several levels: Is the correct information retrieved? Are the sources relevant? Does the answer accurately address the provided context? During the training, you will develop test datasets and use appropriate metrics as well as automated and LLM-based evaluation methods.

    In chunking, documents are divided into smaller sections. If the sections are too large, they may contain too much irrelevant information; if they are too small, important connections may be lost. During the training, you will examine various strategies and their impact on retrieval and response quality.

    Yes. Especially with in-house training, you can use your own documents, knowledge bases, and data sources. This allows you to tailor the pipeline to your company’s specific requirements. Confidential information should only be processed in approved or anonymized environments.

    A prototype requires additional components for production use, such as API deployment, monitoring, access control, versioning, and quality assurance. In addition, data and conceptual drift must be checked regularly. This training session outlines the steps required to move from a notebook prototype to a maintainable solution.

    Scroll to Top