​
Onboarding Agent Tutorial

This tutorial guides you through the inner workings of the Onboarding Agent - the same agent that was automatically created when you first signed up on the Machina Sports platform. While you already have access to this agent in your project, this guide provides a behind-the-scenes look at how it’s built and how you can customize it to better suit your needs. The Onboarding Agent serves as both a practical tool for your development work and an educational example of what’s possible with Machina’s agent architecture.

​
Key Features

Data Synchronization

Automatically sync sports data from multiple providers (NBA, NFL, MLB, NHL, Soccer).

Chat Assistance

Interactive chat interface for answering developer questions during onboarding.

Content Generation

Generate team summaries, game previews, and other content from synchronized data.

Scheduled Tasks

Automate recurring tasks with configurable scheduler agents.

​
Implementation Guide

This section reveals how your existing Onboarding Agent was built and configured. Understanding these components will help you customize the agent or create new agents tailored to your specific needs.

​
1. Installation

The onboarding template that created your agent was installed using the _install.yml configuration: 
setup:
  title: "Machina Studio Onboarding"
  description: AI-powered onboarding for Machina Studio.
  category:
    - special-templates
  estimatedTime: 5 minutes
  features:
    - AI Agent to onboard new developers.
    - AI Reporter to generate onboarding materials.
  integrations:
    - sportradar

​
2. Folder Structure

The template creates a standard folder structure for organizing your data: 
- Catalogue: Game schedules and sports data
- Chat Threads: User messages and conversations
- Embeddings: AI-generated content snippets

​
3. Data Synchronization

Configure data synchronization for different sports leagues:

​
NBA Team Sync

workflow:
  name: "sync-nba-team"
  title: "Sync NBA Team"
  description: "Workflow to sync nba team."
  inputs:
    team_id: "$.get('team_id') or None"
  tasks:
    - type: "connector"
      name: "task-load-team-events"
      connector:
        name: "sportradar-nba"
        command: "get-teams/{team_id}/{data_type}"
        command_attribute:
          team_id: "$.get('team_id')"
          data_type: "'profile.json'"

​
Soccer Team Sync

workflow:
  name: "sync-soccer-team"
  title: "Sync Soccer Team"
  description: "Workflow to sync soccer team."
  inputs:
    team_id: "$.get('team_id') or None"

​
4. Chat Processing

The template includes a thread executor for processing chat messages: 
workflow:
  name: "thread-executor"
  title: "Thread Executor"
  description: "Workflow to execute a thread."
  tasks:
    # Load pending thread
    - type: "document"
      name: "load-pending-thread"
      config:
        action: "search"
        search-limit: 1
        search-vector: false
    
    # Generate response using AI
    - type: "prompt"
      name: "thread-executor-prompt"
      connector:
        name: "machina-ai-fast"
        command: "invoke_prompt"
        model: "llama-3.3-70b-versatile"

​
5. Scheduler Configuration

Set up schedulers to automate recurring tasks: 
agent:
  name: "scheduler-thread"
  title: "Thread Executor Scheduler"
  description: "Scheduler responsible for thread executor."
  context:
    config-frequency: 0.03  # Runs every ~2 seconds
  workflows:
    - name: "thread-executor"
      description: "Workflow Thread Executor"

​
Customization Options

Your Onboarding Agent comes pre-configured with default settings, but you can customize it in numerous ways to better suit your development needs:

​
Data Sources

  • Configure different Sportradar connectors based on your sports focus
  • Add custom data sources by creating new connector configurations

​
AI Models

  • Switch between different language models:
    • OpenAI (gpt-4o) for highest quality responses
    • Groq (llama-3.3-70b-versatile) for faster processing
    • Other models as needed for specific use cases

​
Sync Frequency

  • Adjust scheduler frequencies based on your data freshness requirements
  • Configure different schedules for different data types

​
Content Generation

  • Customize prompt schemas to generate different types of content
  • Modify mapping configurations to transform data for your specific needs

​
Sample Workflows

Your Onboarding Agent already includes these pre-configured workflows that you can study, modify, or use as templates for creating new ones:
  • chat-completions.yml: Process chat messages and generate responses
  • thread-executor.yml: Execute chat threads with context-aware responses
  • sync-nba-team.yml: Synchronize NBA team data
  • sync-soccer-team.yml: Synchronize soccer team data
  • embed-nba.yml: Generate embeddings for NBA content
  • soccer-embeddings.yml: Generate embeddings for soccer content
You can view and edit these workflows directly in your project to see how they’re implemented.

​
Next Steps

Now that you understand how your Onboarding Agent was built, you can:
  • Customize Your Existing Agent: Modify the workflows, connectors, and prompts of your Onboarding Agent to better suit your specific needs
  • Create New Specialized Agents: Use the principles demonstrated in this tutorial to build additional agents for different purposes
  • Explore Advanced Features: Dive deeper into Workflows to understand how to enhance your agent’s capabilities
  • Add Data Sources: Learn about Connectors to integrate additional sports data providers
  • Transform Data: Review Mappings to process and transform data for your specific use cases
  • Enhance Content Generation: See Prompts to improve the quality and relevance of AI-generated content
Your Onboarding Agent is just the beginning of what’s possible with Machina Sports. Use this knowledge to build more sophisticated agents that can automate your sports data workflows and content generation needs.