📚 Learning Flow

AI Agent Tutorial

Quick Start (3 minutes)

System Overview & Core Concepts

Run in Colab (No Code)

Build Your Own AI Agent

FAQ / Troubleshooting

🚀 Go further: create (or customize) your own Agent

Ready to tailor the workflow to your course or research? You can clone our open-source project and modify the Agent to fit your use case.

• Public GitHub: https://github.com/araliadata/AraliaOpenRAG
• Use any Git client (e.g., SourceTree, GitHub Desktop) to pull the repo and explore.

🔧 Where to start (file map)

Use this as your “tour guide” to the codebase:

• graph.py — Orchestrator

  Connects the five Agent stages with LangGraph (fixed execution order).

  Think: the flowchart of your Agent.

• node.py — Agent node logic

  Implements the five core stages (Search → Plan → Decide Filters → Execute → Interpret).

  Think: what each stage does and expects as input/output.

• aralia_tools.py — Tooling layer

  Handles OAuth, dataset search, column/value exploration, and the actual queries to Planets.

  Think: the bridge to Aralia’s data ecosystem.

• schema.py — Data descriptors

  Field types, formats (e.g., year, admin_level_4), and calculations (e.g., avg, sum).

  Think: contracts that keep data aligned.

• prompts.py — Prompt templates

  Templates for dataset filtering, analytics planning, and insight interpretation.

  Think: how the LLM is guided at each stage.

• state.py — LangGraph state

  Shared state across nodes (e.g., question, ai, at, final_response, interpretation_prompt).

  Think: the Agent’s memory during a run.

💡 Quick win: start with the

Interpretation Agent

Adjust the tone, structure, or length of the final summary (e.g., academic abstract vs. executive brief). Run it in the Colab example to see changes instantly.

🛠 Suggested customization path

1. Clone & run the default Agent (use the Colab notebook or your local Python setup).
2. Tweak prompts in prompts.py (e.g., add “limitations” or “policy implications” to the summary).
3. Add validations in analytics_planning_agent to enforce metric/geo/time clarity.
4. Insert a new node (optional) for quality checks or citation formatting (edit graph.py to add the node to the flow).
5. Document your changes so others in your class/research group can reproduce.

🔐 Keep your

Client ID / Client Secret

📚 If you’re new to Python

You don’t need to be a software engineer to get started. These resources cover the essentials:

• Official tutorial: https://docs.python.org/3/tutorial/
• Google Python Class (free): search for “Google Python Class”
• Courses: Coursera – Python for Everybody, edX, Codecademy

Recommended topics to learn first: variables, functions, conditionals, loops, modules, using pip, and running notebooks in Colab.

✅ What “success” looks like

• You can run the baseline Agent and get a reproducible insight (sources + aligned data + summary).
• You can change one thing (e.g., the interpretation style) and see the effect.
• You can explain, in your own words, what each of the five stages does—and where you’d customize next.

When you’re ready, branch off and start shaping the Agent around your class or study—one small improvement at a time.

                                                ← Previous： [**Run in Colab (No Code)**](<https://deciduous-centipede-9d7.notion.site/Run-in-Colab-No-Code-264ddf94fd1480e993c8fe5c71dffd6f>)