faion-ml-engineer（faion 机器学习工程师）

Entry point: /faion-net — invoke this skill for automatic routing to the appropriate domain.

ML Engineer Orchestrator

Communication: User's language. Code: English.

Purpose

Routes AI/ML tasks to specialized sub-skills. Orchestrates LLM integration, RAG, operations, agents, and multimodal AI.

Context Discovery

Auto-Investigation

Check for existing AI/ML setup:

Discovery Questions

Use AskUserQuestion to understand AI/ML requirements.

Q1: AI/ML Goal

question: "What do you want to achieve with AI/ML?"
header: "Goal"
multiSelect: false
options:
  - label: "Use LLM APIs (chat, generation)"
    description: "Integrate OpenAI, Claude, or Gemini"
  - label: "Build RAG system (knowledge base)"
    description: "Search and retrieve from documents"
  - label: "Create AI agent (autonomous tasks)"
    description: "Agent that uses tools and reasons"
  - label: "Fine-tune a model"
    description: "Train model on custom data"
  - label: "Add vision/image/voice"
    description: "Multimodal AI capabilities"

Routing:

• "LLM APIs" → Skill(faion-llm-integration)
• "RAG system" → Skill(faion-rag-engineer)
• "AI agent" → Skill(faion-ai-agents)
• "Fine-tune" → Skill(faion-ml-ops)
• "Multimodal" → Skill(faion-multimodal-ai)

Q2: LLM Provider Preference (if LLM task)

question: "Which LLM provider do you prefer?"
header: "Provider"
multiSelect: false
options:
  - label: "OpenAI (GPT-4)"
    description: "Best general purpose, good tools support"
  - label: "Anthropic (Claude)"
    description: "Best for long context, reasoning, safety"
  - label: "Google (Gemini)"
    description: "Multimodal, 2M context, grounding"
  - label: "Local (Ollama)"
    description: "Privacy, no API costs, offline"
  - label: "Not sure / recommend"
    description: "I'll suggest based on your use case"

Q3: Data Situation (if RAG or fine-tuning)

question: "What data do you have?"
header: "Data"
multiSelect: true
options:
  - label: "Documents (PDF, markdown, text)"
    description: "Unstructured text content"
  - label: "Structured data (database, CSV)"
    description: "Tabular or relational data"
  - label: "Code repositories"
    description: "Source code to search/understand"
  - label: "Conversation logs"
    description: "Chat history, support tickets"

Routing:

• "Documents" → RAG with chunking strategies
• "Structured data" → Text-to-SQL or structured RAG
• "Code repos" → Code embeddings, AST-aware chunking
• "Conversations" → Fine-tuning dataset prep

Q4: Deployment Requirements

question: "How will this be deployed?"
header: "Deploy"
multiSelect: false
options:
  - label: "API endpoint (backend service)"
    description: "Part of web application"
  - label: "CLI tool"
    description: "Command-line interface"
  - label: "Batch processing"
    description: "Process data in bulk"
  - label: "Real-time/streaming"
    description: "Live interactions, low latency"

Context impact:

• "API endpoint" → Async patterns, rate limiting, caching
• "CLI tool" → Simple integration, local models option
• "Batch processing" → Cost optimization, parallel processing
• "Real-time" → Streaming responses, edge deployment

Sub-Skills (5)

Total: 101 methodologies

Routing Logic

Execution Protocol

When a task arrives:

1. Analyze task intent
2. Select appropriate sub-skill (use routing table above)
3. Invoke sub-skill with Skill tool
4. Return results to caller

Quick Reference

Related Skills

*ML Engineer Orchestrator v2.0* *5 Sub-Skills | 101 Total Methodologies*