I'm a backend engineer working with Java / Kotlin / Spring Boot and open-source contributor on spring-projects/spring-ai, currently exploring schema generation, transport-layer plumbing, and provider documentation consistency.

An Application Framework for AI Engineering

• Problem: Spring AI inlined parameter schemas under properties.<param>, but their generated $refs still pointed to root-level $defs. For recursive parameter types, this produced an unresolvable $ref in the tool input schema.
• Fix: Hoist each parameter schema's $defs up to the wrapper schema root before inlining. Reuse structurally equal definitions, rename simple-name collisions, and rewrite peer $refs inside the hoisted schema.
• Scope: Applied symmetrically to both JsonSchemaGenerator (spring-ai-model) and McpJsonSchemaGenerator (mcp/mcp-annotations).
• Tests: Added regression coverage for transitive recursive parameter schemas, duplicate equal definitions, and colliding simple-name definitions with ref rewrites.

Follow-up to #6036, completing the corrections started in commit 192eb5c.

• Problem: After the openai-java SDK migration, OpenAI-compatible base URLs and several *-path properties were inconsistent across 8 documentation files (DMR, Ollama, Groq, Mistral AI, Embeddings, Speech, Transcriptions, Moderation). Some still missed /v1; some still documented properties that had been dropped from the SDK.
• Fix: Aligned all OpenAI-compatible endpoints to include /v1 (e.g. http://localhost:12434/engines/v1, https://api.groq.com/openai/v1), removed table rows and sections for properties dropped by the SDK migration, and verified AsciiDoc structure stayed intact.
• Outcome: Merged into milestone 2.0.0-M7 (d7f3c6c).