Concerto for Java and AI - Building Production-Ready LLM Applications

Imagine you're a music composer struggling to find inspiration for a pivotal movie scene. Then, you remember you’re also a software engineer, and the solution becomes suddenly obvious. Join me in this session, where I'll demonstrate how I enhanced my music composition process by harnessing the power of Java and AI.

This presentation will discuss the core architectural patterns for introducing AI capabilities into an existing software system, exploring concepts like templated prompts, agent tools, and integration of external knowledge. The Java ecosystem is getting more and more capabilities for building AI applications. But are they ready for production? Are there any gaps?

Throughout the session, I’ll build a "composer assistant" application using Java and an AI Orchestrator (LangChain4J, Spring AI) to showcase how to make an LLM application production-ready. Is the developer experience affected when working locally with language models? How is observability different when it comes to tokens? Can we ensure resilience across the many integrations orchestrated by the AI? What strategies are available for deploying LLM applications?

In a final twist, you’ll choose which movie scene to score, and I’ll compose and perform the music live for it, supported by AI. Will it meet the mark? There’s only one way to find out: join me in exploring the practical side of AI applications, where Java and Generative AI offer tangible solutions to real-world use cases. Aaaaand action!

Transcript

1. Thomas Vitale GOTO Copenhagen Oct 2nd, 2024 Concerto for Java and AI Building Production-Ready LLM Applications @vitalethomas
2. Systematic • Software Engineer • CNCF Ambassador, Oracle ACE Pro, Testcontainers Community Champion • Author of “Cloud Native Spring in Action” (Manning). • OSS contributor (Java, Spring, Cloud Native Technologies) Thomas Vitale thomasvitale.com @vitalethomas
3. @vitalethomas LLM RAG Prompting Embeddings Vector Stores Hallucinations Agents Generative AI Ph. Francisco Emilio Diaz
4. One Buzzword To Rule Them All @vitalethomas
5. The WHY Factor @vitalethomas
6. The WHY Factor @vitalethomas What problem does it solve? How ready is it for production? Yeah, but how about the DevEx?
7. Large Language Models @vitalethomas
8. Machine Learning Subset of Arti fi cial Intelligence @vitalethomas Model Training Model Fine-Tuning Model Inference ML Engineers Platform/Infrastructure Platform Engineers HTTP API Application Developer
9. If you like it, you should put an API on it @vitalethomas
10. Model Inference via HTTP APIs @vitalethomas Application Model Inference Service Do you wanna build a snowman? HTTP Application Database Service DELETE * FROM HYPE; JDBC
11. Text Classification @vitalethomas
12. Text Classification @vitalethomas Application Model Service Classify HARMONY
13. LLM Security Risks (1) OWASP Top 10 for LLM Prompt Injection Model Denial of Service Sensitive Information Disclosure @vitalethomas OWASP Top 10 LLM Applications and Generative AI https://genai.owasp.org/
14. Resilience LLM Applications @vitalethomas Resilience4J
15. Observability Distributed Tracing Request Rate Errors Duration Prompt Content Token Usage Context Window @vitalethomas
16. Semantic Search @vitalethomas
17. Semantic Search @vitalethomas From Keywords to Meaning Application Melancholic Embedding Model Melancholic [42…] LIKE ‘%melancholic%' Database Vector Store [42…]
18. Data Ingestion LLM Applications @vitalethomas Document Reader Document Transformer Document Writer JobRunr
19. Question Answering @vitalethomas
20. Question Answering with Docs @vitalethomas Retrieval Augmented Generation Application Melancholic instrument? Embedding Model Melancholic instrument? [42…] Get Similar Documents Vector Database Model Service Question + Similar Documents
21. One Does Not Simply Test LLM Applications @vitalethomas
22. Structured Data Extraction @vitalethomas
23. Structured Data Extraction @vitalethomas From Text to JSON Application Text Text to Structured JSON Model Service Database Save Structured JSON
24. LLM Security Risks (2) OWASP Top 10 for LLM Insecure Output Handling Excessive Agency Insecure Plugin Design @vitalethomas OWASP Top 10 LLM Applications and Generative AI https://genai.owasp.org/
25. Hallucinations @vitalethomas
26. Data Validation JSON Schema Humans in the Loop Optional Values @vitalethomas Mitigating hallucination risks
27. Speech Transcription @vitalethomas
28. Speech Transcription @vitalethomas From Speech to Text Application Audio Audio to Text Audio Model Chat Model Text to Structured JSON
29. Privacy @vitalethomas
30. Build & Deploy Cloud Native Buildpacks Kubernetes Service Binding Native Executables with GraalVM @vitalethomas Going to Production
31. Service Bindings for Spring AI @vitalethomas
32. @vitalethomas Ph. Francisco Emilio Diaz Composer Assistant https://github.com/ThomasVitale/concerto-for-java-and-ai
33. Thomas Vitale @vitalethomas thomasvitale.com Concerto for Java and AI Building Production-Ready LLM Applications