Enhance AI Fitness Microservices Platform

I already have the first version of a fitness-tracking platform running on Spring Boot microservices, a React front-end, Docker containers, a Keycloak OAuth2 realm, and an Apache Kafka event bus. The foundations are solid, but I now want to push the system further so it truly shines with its AI-driven features.

What the next milestone must achieve
• The AI service should deliver personalised workout plans, provide real-time feedback during exercises, and update a daily activity log—all generated on the fly through Google Gemini or OpenAI models.
• The User service needs to capture and persist richer profile data—specifically each member’s fitness goals, health metrics, and exercise preferences—so the AI has the context it needs.
• Kafka topics for activity-tracking updates and AI-recommendation requests must be streamlined, with clear schemas and robust consumer logic to prevent message loss or duplication.
• Everything continues to flow through the existing Spring Cloud API gateway and config server, secured by Keycloak. Any new endpoints, scopes, or roles you introduce should respect that guard-rail.
• The React client should surface the new recommendations live, without page refreshes; a lightweight WebSocket layer or Kafka-to-Web front channel is fine as long as latency stays low.

Acceptance criteria
1. A demo workflow showing a new user setting goals and immediately receiving an AI-generated plan, followed by live exercise feedback and an updated daily summary.
2. Unit and integration tests covering the new Kafka topics and AI endpoints (minimum 80 % coverage for touched code).
3. Docker-Compose or Kubernetes manifests updated so a single command spins up the entire stack with sample data.
4. A short README outlining any environment variables, model keys, or Keycloak realm changes.

If you’re comfortable juggling Spring Boot, React, Kafka, Keycloak, Docker, and one of the major LLM APIs, let’s refine this product together and get users moving smarter.