Bringing It All Together, Documenting Your Own System Design

Design conversation memory for multi-turn dialogues using Spring AI Advisors (MessageChatMemoryAdvisor, PromptChatMemoryAdvisor)

Full component diagram of your Spring AI capstone + System Design of an Online Chess Website (WebSocket real-time state sync, matchmaking engine, game state persistence, rating calculation)

Design handling for: empty vector retrieval results, tool execution failure, prompt injection attempts, context window overflow

Trace request end-to-end on paper; document failure modes (provider downtime, slow vector search, DB connection pool exhaustion) + System Design of Google Docs Collaborative Real-Time Editor (Operational Transformation - OT vs CRDT, real-time conflict resolution, snapshotting, WebSocket synchronizer)

Define quantitative evaluation rubric for the integrated capstone: accuracy of tool selection, RAG grounding, response latency, token cost

Combine RAG, tool calling, conversation history memory, and rate limiting in single production Spring Boot application.

Integrate JWT authentication into AI endpoints; persist conversation memory in PostgreSQL; add Spring Cloud Gateway rate limiting

Integrate Micrometer and OpenTelemetry for LLM token usage tracking, latency spans, and error monitoring.

Wire all components together: Auth + Memory + RAG + Tool Calling + MCP Server; run automated end-to-end integration tests on GitHub