Orchestrate Production AI/ML Workflows

This workflow bundle orchestrates a comprehensive AI/ML build across seven phases, each naming specific skills to invoke: Phase 1 AI Application Design (ai-product, ai-engineer, ai-agents-architect, llm-app-patterns - define use cases, model choice, architecture, success metrics); Phase 2 LLM Integration (ai-assistant dev, LangChain agents, prompt optimization, Gemini API - provider selection, prompt templates, streaming, error handling); Phase 3 RAG Implementation (rag-engineer, embedding-strategies, vector-database-engineer, hybrid-search-implementation - data pipeline, chunking, retrieval, reranking, caching); Phase 4 AI Agent Development (autonomous-agents, crewai, langgraph, multi-agent-patterns, computer-use-agents - roles, tool integration, memory, orchestration, human-in-the-loop); Phase 5 ML Pipeline Development (ml-engineer, mlops-engineer, data-engineer - training, evaluation, model registry, deployment); Phase 6 AI Observability (langfuse, manifest, llm-evaluation - tracing, logging, cost tracking, alerts); and Phase 7 AI Security (prompt-engineering security, SAST scanning - input validation, output filtering, rate limiting, access controls, audit logging).

It includes checklists for LLM integration (API keys secured, rate limiting, streaming, token tracking), RAG systems (pipeline working, embeddings generated, retrieval accuracy tested, caching implemented), AI agents (roles defined, tools integrated, memory working, orchestration tested), and observability (tracing, metrics, evaluation, alerts, dashboards), plus overall quality gates before considering the work done.

Use this as the overarching roadmap when building LLM-powered applications, implementing RAG, creating AI agents, developing ML pipelines, adding AI features to an existing application, or setting up AI observability - anywhere multiple AI-related concerns need to be sequenced together.

Not a substitute for the specialist skills it names - each phase delegates the actual implementation work to a dedicated skill (rag-engineer, crewai, langgraph, ml-engineer, langfuse, etc.). Use this to know what to build in what order and which skill handles each piece, not as the implementation itself.

Inputs: the target AI feature or system scope (LLM app, RAG, agents, ML pipeline, or a combination) and the constraints/success metrics for each.

Outputs: a full AI system spanning LLM integration, RAG (if applicable), agents (if applicable), ML pipelines (if applicable), observability, and security controls - each phase checked against its own quality checklist before moving to the next.

Delegates to ai-product, ai-engineer, ai-agents-architect, rag-engineer, vector-database-engineer, crewai, langgraph, ml-engineer, mlops-engineer, langfuse, and prompt-engineering security skills, among others; related bundles include development, database, cloud-devops, and testing-qa.

Teams planning a comprehensive AI/ML build spanning multiple concerns - LLM integration, RAG, agents, ML pipelines, observability, and security - who need a sequenced roadmap naming the right specialist skill at each phase.