Spring AI 2.0 · OpenAI · Gemini · Ollama
Evidence enters.
Action is earned.
Build AI systems that retrieve the right evidence, choose bounded tools, validate every action, and survive real production failure.
INTENT → RETRIEVAL → EVIDENCE → MODEL → TOOL → VALIDATION → ACTION → EVAL
24 detailed chapters
145 concepts
48 decision drills
78 scored questions
From token to production
Follow one decision through the whole system.
Connect Spring AI, models, embeddings, retrieval, tools, agents, tuning, evaluation, security, and operations.- 01UnderstandModels, tokens, Spring AI 2, prompts, outputs, and provider routingCh 00–04
- 02ConnectOpenAI, Gemini, Ollama, embeddings, semantic search, and vector storesCh 05–10
- 03GroundIngestion, basic and advanced RAG, memory, tools, and MCPCh 11–15
- 04ControlAgents, approvals, fine-tuning, local adapters, evaluation, and securityCh 16–21
- 05OperateProduction architecture and the complete ClaimsPilot implementationCh 22–23
- 01Intent is scopedidentity and authority are known
- 02Evidence retrieveshybrid search finds sources
- 03Model proposesreasoning stays inside boundaries
- 04Tools are checkedserver policy authorizes calls
- 05Evaluation provesclaims, cost, latency, and safety pass
Real-life AI architecture
ClaimsPilot
Help insurance adjusters review evidence without allowing a model to decide or pay a claim.
Design multimodal ingestion, hybrid retrieval, policy-date filters, citations, provider routing, read-only tools, structured drafts, human approval, evaluations, security, canaries, and rollback.
Open the complete scenario ↗PACKETRETRIEVALMODELAPPROVALparse · authorize · retrieve · reason · validate · approve
Recall, decide, explain
Build judgment—not framework trivia.
A framework is a set of ownership rules.