Spring AI

🤖 Spring AI (+ LangChain4j)

Checklist

• ChatClient fluent API, system/user messages, prompt templates
• Model portability: Anthropic, OpenAI, Gemini, Ollama (local), Bedrock
• Streaming responses (Flux → SSE)
• Structured output → Java records
• Tool calling (@Tool methods) and returning tool results
• Advisors: chat memory, RAG (QuestionAnswerAdvisor), logging, safeguards
• VectorStore abstraction: pgvector, Redis, etc.; ETL pipeline (DocumentReader → splitter → embed → store)
• MCP: Spring AI MCP server & client starters ⭐ → MCP
• Observability: Micrometer metrics for tokens and latency
• Evaluation helpers (relevancy/fact-checking evaluators)
• Compare with LangChain4j (AI Services, agents) and know when you’d pick each

🧪 Labs (🟢 warm-up → 🟡 core → 🔴 hard → ⚫ boss)

• 🟢 ChatClient with streaming + structured output into records
• 🟡 @Tool methods + advisors (memory, logging); compare with LangChain4j AI Services
• 🔴 An ETL pipeline into pgvector + QuestionAnswerAdvisor, then replace it with your own hybrid retriever
• 🔴 Orbit MCP server with the Spring AI MCP starter (tools + resources + OAuth)
• ⚫ The MCP client side: connect external MCP servers as Orbit tools with scopes

🧠 Cognitive tasks

• Framework vs hand-rolled: what does Spring AI’s tool calling do that your Go loop does by hand?

🛰️ Orbit integration

• orbit-knowledge (RAG) and orbit-tools (MCP server/client) → Orbit v4 - Agents, RAG & MCP

Go deeper

⚙️ LLM Inference Internals · 🧩 AI & Agent Patterns · 🤖 MCP

Resources

• docs.spring.io/spring-ai ⭐ · Spring AI examples repo · Dan Vega + Josh Long videos on Spring AI · docs.langchain4j.dev Related: AI Engineering Roadmap