RAG Tutorial Series — Java + Spring AI + Angular

🇩🇪 Deutsche Version

A four-part, progressively-built tutorial series that teaches Retrieval-Augmented Generation (RAG) from first principles to a production deployment. Each part is a separate, fully self-contained, independently runnable Maven project — you can clone (or copy) a single part's folder and have everything you need. Later parts are supersets of earlier parts.

Every part ships with an English (README.md) and a German (README.de.md) walkthrough.

What is RAG?

Large language models only know what was in their training data. RAG lets a model answer questions about your documents by combining three steps at question time:

                 ┌─────────────────────────────────────────────────────┐
                 │                     RAG at question time            │
                 │                                                     │
  question ───▶  │  1. RETRIEVAL      similarity-search the vector     │
                 │                    store for the most relevant      │
                 │                    document chunks                  │
                 │  2. AUGMENTATION   put those chunks into the        │
                 │                    prompt as context                │
                 │  3. GENERATION     let the LLM answer *grounded*    │
                 │                    in that context                  │
                 └─────────────────────────────────────────────────────┘

Before any question can be answered, documents must go through the ingestion pipeline once:

  file / URL ──▶ extract text ──▶ chunk ──▶ embed ──▶ store
               (Tika / Jsoup)   (token     (embedding  (vector
                                 splitter)   model)      store)

• Embedding — a vector of floats that captures the meaning of a text; texts with similar meaning get vectors that are close together.
• Vector store — a database optimized for "give me the N stored chunks whose vectors are closest to this query vector".
• Chunking — splitting long documents into pieces small enough to embed well and to fit several of them into one prompt.

The four parts

Part	Folder	What it adds
1	rag-tutorial-01-basics	The smallest working RAG loop: Spring Boot + in-memory SimpleVectorStore + Azure OpenAI, file/URL ingestion (Tika + Jsoup), Angular dashboard
2	rag-tutorial-02-llms	Swappable LLM providers (azure / openai / ollama — incl. fully offline local Mistral) and a persistent pgvector store, all via Spring profiles + docker-compose
3	rag-tutorial-03-connectors	Optional, off-by-default connectors: Confluence ingestion, Microsoft Teams outgoing webhook, Slack Events API bot — each with signature verification and async Q&A re-embedding
4	rag-tutorial-04-full	Everything combined + production packaging: Dockerfiles for backend and frontend (nginx), and a Helm chart with optional Postgres/pgvector and Ollama deployments

Tech stack (fixed across all parts)

Building block	Choice
Backend language	Java 17
Backend framework	Spring Boot 3.4 + Spring AI 1.0
LLM / Embeddings	Azure OpenAI, OpenAI API, local Mistral via Ollama (Docker) — swappable via Spring profile (from Part 2)
Vector store	In-memory SimpleVectorStore and PostgreSQL + pgvector — swappable via Spring profile (from Part 2)
Document sources	Local files (PDF/DOCX/TXT via Apache Tika) and web URLs (Jsoup); Confluence from Part 3
Frontend	Angular 18 standalone components, plain CSS, no UI framework

Note on versions: the original series plan said "Spring Boot 3.3", but Spring AI 1.0 requires Spring Boot 3.4.x — all parts therefore use Spring Boot 3.4.

Blog walkthroughs

Each part has a companion article on halvic.ch that explains the design decisions behind the code (English and German):

Part	English	Deutsch
1	Build a RAG app with Spring AI	RAG-App mit Spring AI bauen
2	Swap LLMs without code changes	LLM austauschen ohne Code-Änderung
3	RAG connectors: Confluence, Teams, Slack	RAG-Konnektoren: Confluence, Teams, Slack
4	Deploy RAG with Docker and Helm	RAG-Deployment mit Docker und Helm

How to use this series

Start with Part 1 and read its README top to bottom — it explains every concept. Each later README repeats the fundamentals so it also works standalone, then focuses on what is new. If you only want the final result, jump straight to Part 4.