Source Code and Manuscript Files for the October 2026 edition of "Practical Artificial Intelligence With Java"
The new October 2026 edition is largely a refresh fro new library and tool versions and new LLM model versions.
Leanpub Link for latest edition · read the book free online
This book is a combination of
- new coverage of LLMs
- new coverage of deep learning
- new material: creating and using knowledge graphs
- examples from my discontinued book "Power Java": anomaly detection, linked data, using DBPedia, OpenNLP, and web scraping
- examples from the original editions of this book: genetic algorithms and search algorithms
- a few examples updated from my discontinued book "Practical Semantic Web and Linked Data Applications, Java Edition"
manuscript/— the book chapters as Markdown.manuscript/Book.txtlists the chapter order used by Leanpub, andSOURCE_CODE_MAP.mdmaps each chapter to its example project.source-code/— one Maven project per chapter, plus the sharedprompts/directory of prompt templates.
| Tool | Version | Notes |
|---|---|---|
| JDK | 21 or newer | Every pom.xml compiles with --release 21; verified with Temurin 21 and 27. The optional formatter targets need a JDK 21 toolchain — see the note in section 5. |
| Maven | 3.9 or newer | Resolves dependencies and runs the examples. |
| GNU Make | any recent version | Optional convenience wrapper; every project also builds with plain Maven. |
| Ollama | current release | Only needed for the two local-LLM projects, ollama-llm-client and langchain4j-ollama. |
Confirm your toolchain before starting:
java -version
mvn -versionThe cloud LLM examples read credentials from environment variables. Nothing needs to be set to build the code or to run the fully offline examples.
| Variable | Used by | Purpose |
|---|---|---|
OPENAI_API_KEY |
source-code/openai-llm-client |
OpenAI key (read automatically by the official OpenAI Java SDK). |
GOOGLE_API_KEY |
source-code/gemini-llm-client |
Gemini API key from Google AI Studio. |
GEMINI_API_KEY |
source-code/AgentScope_gemini |
Gemini key used by the AgentScope ReActAgent examples. |
OLLAMA_BASE_URL |
source-code/langchain4j-ollama |
Optional; overrides the default http://localhost:11434. |
GEONAMES |
source-code/info_gathering |
Optional; your free GeoNames account name for the GeoNames web-service example. |
For example:
export OPENAI_API_KEY=sk-...
export GOOGLE_API_KEY=...
export GEMINI_API_KEY=...cd source-code
make installmake install runs mvn dependency:go-offline and then compiles (or packages) each project. This primes your local ~/.m2 repository so that later builds can run with Maven's offline flag (-o), and it installs the shared libraries (ner_dbpedia, semantic_web_apache_jena) before the projects that depend on them (kgc, kgn). This step needs network access the first time you run it. Setup deliberately does not run the formatter — make fmt rewrites files, so it stays a separate, optional target.
cd source-code
make checkmake check compiles the main and test sources of all 15 projects (mvn -o -q test-compile) without running any tests. After make install has completed, this is fully offline.
Run one project:
cd source-code/openai-llm-client
make run # or: mvn test -qRun every project's tests:
cd source-code
make testmake test is the one target that is not run in Maven's offline mode. Surefire resolves its JUnit Platform provider at test time, and mvn dependency:go-offline cannot pre-cache it, so the first make test fetches that provider; after that it is cached like everything else. make check (compile only) stays fully offline.
Other top-level targets: make fmt (format with Spotless), make fmt-check (verify formatting), make verify (fmt-check + check), and make clean.
One toolchain caveat: the google-java-format version pinned by these projects does not run on JDK 27, so make fmt, make fmt-check, and make verify cannot run there. Those targets need a JDK 21 toolchain, and even then make fmt-check reports drift in the checked-in sources (the code is written in 4-space style while the config asks for Google's 2-space style), so make fmt will rewrite files rather than confirm them. Treat formatting as optional and use make check / make test as the real gates; those work on newer JDKs.
The table below describes what make test / make run needs. Note that make check (compile only) is offline for every project once make install has run.
| Project | Offline? | Requirements for tests / examples |
|---|---|---|
anomaly_detection |
Yes | — |
genetic-algorithms |
Yes | — |
neuralnetworks |
Yes | — (starts a Swing GUI) |
nlp |
Yes | — |
ner_dbpedia |
Yes | Install before kgc / kgn |
kgc |
Yes | Depends on ner_dbpedia |
search |
Yes | — (starts a Swing GUI) |
ollama-llm-client |
Locally | A running Ollama server (v0.35.1+) with a small model pulled, e.g. ollama pull qwen3.5:2b |
langchain4j-ollama |
Locally | A running Ollama server with a small model pulled |
openai-llm-client |
No | OPENAI_API_KEY + network access |
gemini-llm-client |
No | GOOGLE_API_KEY + network access |
AgentScope_gemini |
No | GEMINI_API_KEY + network access |
info_gathering |
No | Network access (web scraping); GEONAMES for the GeoNames example |
semantic_web_apache_jena |
Partly | Network access for the remote DBpedia SPARQL queries; local queries are offline |
kgn |
Partly | Network access for remote SPARQL; build ner_dbpedia and semantic_web_apache_jena first |
"Locally" means no cloud API key is required, but the examples still need the Ollama server running on your machine.
Model identifiers, provider APIs, and library versions change quickly. The LLM chapters and examples were audited in October 2026; where a chapter recommends a specific model or dependency version, treat it as a starting point and check the provider's current documentation before relying on it. The dependency versions that actually matter for building are the ones in each project's pom.xml.