RAG training
Retrieval-Augmented Generation that actually scales
RAG is a demo away, and a different fight in production. In this training you learn the full spectrum: chunking strategies, embeddings, hybrid search, re-ranking and the evaluation approach most teams skip until it bites them.
What you learn
- Chunking strategies and their impact on answer quality
- Choosing and benchmarking embedding models on your own data
- Vector stores next to hybrid search (BM25 + vector) and when each wins
- Re-ranking with cross-encoders and LLM rerankers
- Multi-hop, agentic and GraphRAG patterns
- Evaluation with RAGAS, custom test sets and human review
- Production: caching, cost, observability and reindex strategy
Who it's for
- 01Developers and ML consultants making knowledge bases searchable
- 02Data consultants setting up ETL for RAG pipelines
- 03Product owners who want to understand the quality of their RAG systems
Programme
RAG fundamentals
Pipeline, terminology and where quality really comes from.
Chunking and embeddings
Strategies with and without structure, and how to benchmark models.
Search and re-ranking
Hybrid search, BM25, vector and re-ranking models.
Evaluation
RAGAS, test sets, regressions and the gap between 'feels good' and 'is good'.
Production
Caching, cost, observability and when to reindex.
What you take away
What you take away
- A RAG pipeline on your own dataset with clear metrics
- Evaluation approach that catches regressions before release
- Decision tree for vector versus hybrid versus GraphRAG
Prerequisites
- Python at reading level
- Prior experience with a vector store or search engine helps
Frequently asked questions
Which tools do you use?+
We show several stacks — including LlamaIndex, LangChain, pgvector and Qdrant — and focus on the patterns rather than a single framework.
What does the training cost?+
Pricing on request. We'll send a tailored quote that includes course materials, recordings and 30 days of async support afterwards. For in-company sessions you pay one flat fee, regardless of headcount up to 12 people.
Where does the training take place?+
By default at our office in Utrecht or in-company at your own location. Fully remote is fine too; we then split the day into two half-day blocks.
Do I receive a certificate?+
Yes. Afterwards you receive a certificate of attendance listing the covered topics and hours, usable for CPD points and internal records.
Schedule this training
Discuss your learning goals during a no-obligation intake. Afterwards, you receive a tailored programme and written quote.
Other trainings
Decision models training
Explicit decision logic alongside generative AI.
Data, decision models & evaluationAgent evaluation & observability training
Know whether your AI Agent works — including after a change.
Low-code & automationn8n training
Workflow automation that survives production