AI Learning Guide

Aralia AI SDK

AI Agent Tutorial

Quick Start

System Overview & Core Concepts

Prerequisite

Experiment in Colab

Custom AI Workflow

🔌DataPlanet API

Summary

In this chapter, we’ll roll up our sleeves and break down how an AI-powered workflow actually works.

Our AI toolkit is written in Python. Normally, running Python requires lots of setup (interpreters, packages, permissions, etc.). In this tutorial we use Colab—a Google service that prepares the runtime for you. All you need to do is open a browser → connect to the cloud → press “Run” step by step.

colab.google

This way you avoid installation headaches and don’t need to write code, which is ideal for classrooms and first-time users.

In the Colab tutorial, there are three main sections:

• System initialization: Guidance on initializing system tools and the chosen LLM provider.
• Kickoff flow: Apply the Aralia AI Tools to LangGraph and run a sample question to inspect results.
• Deep dive: Detailed explanations of how the entire flow operates.

Use Colab to Understand the AI Workflow

1. Open the tutorial:

   https://colab.research.google.com/drive/1GPllPWemvxodlIkomXqeCWXZon4_9UTF

   or the shared Notebook

2. Click Connect (top-right or top-left) to connect to a hosted runtime (Colab will start the machine for you).

   

3. Following the Notebook prompts, paste your Secrets/Keys:

   • Client ID / Client Secret (see Apply for an Aralia Account & Keys)

   • LLM API Key (see LLM API Key). You can start with Gemini, then try OpenAI/Anthropic later.

     

4. Click the ▶ (Run icon) on each cell in order:

   • Install & import

   • Initialize (OAuth sign-in, set Planet)

   • Paste your complete question into the code (you can start with “Malaysia GDP × Gini”)

     

5. You’ll see:

   • A list of data sources found
   • An LLM-generated insight summary (about 300 words)

Dive Deeper into the AI SDK Workflow

Aralia OpenRAG SDK

The Aralia OpenRAG SDK is publicly available on GitHub. You can clone it with any Git tool (e.g., SourceTree) to read and practice:

https://github.com/araliadata/AraliaOpenRAG

https://github.com/araliadata/AraliaOpenRAG

The SDK uses LangGraph as the orchestration backbone to implement a data-analytics workflow for AI. LangGraph provides state management and workflow coordination; the pipeline handles user questions and produces data-driven insights from the Aralia data platform.

Unlike traditional SDKs, you don’t need to understand every internal detail. The LLM inside LangGraph helps “understand” how to use the SDK and automatically decides how to apply it (e.g., which parameters to pass, how to parse results).

What developers need to do:

1. Use LangGraph to define the execution order.
2. Expose the SDK’s Tools to LangGraph.

The system then follows the steps defined in LangGraph to use and call the Tools—no need for users to know how to call each Tool; they only need to understand what each Tool does.

For the LangGraph definition, see the example below.

Graph Definition

The primary workflow is implemented with LangGraph’s StateGraph:

https://github.com/araliadata/AraliaOpenRAG/blob/main/core/graph.py#L59-L85

How Does the AI Retrieve Insights?

You enter a complete question, and the system uses the LLM’s understanding to determine what you need. It then calls five tool nodes in sequence:

Find datasets → Plan fields/charts → Decide query filters → Execute query → Generate insights.

The entire flow is built on RAG (retrieve first, then generate) and vector semantic search.

graph LR
      A[Question] --> B["Aralia<br/>Search<br/>Node"]
      B --> C["Analytics<br/>Planning<br/>Node"]
      C --> D["Filter<br/>Decision<br/>Node"]
      D --> E["Analytics<br/>Execution<br/>Node"]
      E --> F["<br/>Interpretation<br/>Node"]
      F --> G[Answer]

subgraph "Workflow"
B
C
D
E
F
end

style A fill:#e1f5fe
style G fill:#e8f5e8
style B fill:#fff3e0
style C fill:#fff3e0
style D fill:#fff3e0
style E fill:#fff3e0
style F fill:#fff3e0

The Five Tool Nodes (What do they do? What are the I/O?)

1. Aralia Search Node (find datasets)
   • Uses semantic retrieval across multiple Planets (e.g., SDG, UN) to find the most relevant datasets.
   • Output: Candidate datasets + field information.
   • 🔍 Search other Planets or browse published Galaxy datasets
2. Analytics Planning Node (plan analytics)
   • Decides x-axis, y-metrics (incl. calculations like average/sum), filters (year/geography), and field formats (e.g., year, admin_level_4).
   • 🔍 Query datasets
   • 🔍 Query Virtual Variable list
3. Filter Decision Node (finalize filters)
   • Fills in year/geography, etc., to produce a query-ready parameter set.
   • 🔍 Query filter options
4. Analytics Execution Node (execute query)
5. Interpretation Node (generate insights)

Q: API Key / OAuth errors?

A: Check that your keys are correct, not expired, and have sufficient quota.