Muhammad Yasin Khan

Posted on May 14

🌍 GeoHazard AI — Building a Multi-Agent Geological & Climate Risk System with Google ADK

#agents #buildmultiagents #gemini #adk

Education Track: Build Multi-Agent Systems with ADK

This post is my submission for "DEV Education Track: Build Multi-Agent Systems with ADK" (https://dev.to/deved/build-multi-agent-systems).

What I Built

Geological hazards such as landslides are rarely caused by a single factor.
They emerge from interaction between terrain conditions and climate dynamics.

In this project, I built an educational Multi-Agent AI system using Google Agent Development Kit (ADK) that mirrors how real scientists collaborate.

Instead of using one large AI model, the system separates expertise into independent agents:

A Hazard Agent that evaluates geological instability
A Climate Agent that analyzes environmental forcing
A combined GeoHazard reasoning workflow

The goal was to demonstrate how multi-agent architecture improves scientific reasoning, transparency, and modular AI design.

This project serves as a learning example for students, developers, and researchers interested in:

✅ Multi-Agent Systems
✅ Scientific AI workflows
✅ Climate & Earth Science applications
✅ Agent specialization using ADK

Cloud Run Embed

⚠️ Deployment Note

The agents were successfully executed during development using Google ADK.
Later, access to the original cloud execution environment became restricted due to project permission and billing limitations.
(Project Name: Geohazard Agents

Project number: 322609188016

Project ID: geohazard-agents

Dashboard
Cloud Hub
https://console.cloud.google.com/welcome?project=geohazard-agents)

Since this submission focuses on the Educational Track, the project is presented as a reproducible learning system rather than a live deployment.

Readers can deploy the system themselves by:

Creating a Google Cloud Project
Enabling Vertex AI
Installing Google ADK
Running the agent scripts provided below

This approach emphasizes learning and architecture understanding rather than infrastructure setup.

Your Agents

🪨 Hazard Agent — Geological Specialist

Role:
Evaluates terrain instability and landslide susceptibility.

Responsibilities

Analyze slope conditions
Identify terrain risk factors
Classify hazard level

Design Principle

The agent is intentionally restricted:

«It does NOT analyze climate data.»

This teaches role separation, a core concept in Multi-Agent Systems.

Example tool logic:

@hazard_agent.tool
def landslide_tool(slope: float, rainfall: float):
if slope > 30 and rainfall > 100:
return "High Landslide Risk"
return "Moderate Risk"

🌦️ Climate Agent — Environmental Analyst

Role:
Studies climate influence on hazard amplification.

Responsibilities

Evaluate rainfall intensity
Assess temperature anomalies
Identify environmental stress factors

Design Principle

The climate agent avoids geological interpretation, ensuring:

✅ cleaner reasoning
✅ modular upgrades
✅ independent agent learning

🌐 GeoHazard Reasoning (Multi-Agent Collaboration)

The system demonstrates how agents collaborate:

Hazard Agent evaluates terrain
Climate Agent analyzes environmental forcing
Combined reasoning produces final hazard understanding

This mirrors real interdisciplinary scientific workflows.

🔁 Workflow Overview

Environmental Inputs:

Slope angle
Rainfall conditions
Climate anomalies

Processing Flow:

Environmental Data
↓
Hazard Agent → Geological Risk
↓
Climate Agent → Climate Influence
↓
Integrated GeoHazard Assessment

Example Educational Output:

Hazard Level: High
Climate Risk: Elevated
Integrated GeoHazard Risk: High
Recommended Action: Monitoring and mitigation planning required.

Key Learnings

🧠 1. Multi-Agent AI Thinks Like Scientists

Real experts specialize.

Separating AI agents dramatically improved reasoning clarity compared to a single model handling everything.

⚙️ 2. Constraints Improve Intelligence

Limiting agents to defined responsibilities prevented hallucinated conclusions and produced more structured outputs.

🌍 3. AI for Scientific Discovery Is Practical

Multi-agent systems can assist:

Landslide early warning
Climate risk assessment
Infrastructure planning
Environmental decision support

🚧 4. Biggest Challenge

The most challenging aspect was understanding how to coordinate agents instead of scaling a single prompt.

This project shifted my perspective from:

«Prompt Engineering → System Engineering»

🎓 Educational Value

This project demonstrates that building intelligent systems is not only about powerful models — it is about designing collaborating AI experts.

Students following this tutorial learn:

Agent architecture design
Role specialization
Scientific reasoning workflows
Practical use of Google ADK

👨‍🔬 Author Background

I am a Geologist exploring how Artificial Intelligence can support Earth science research, climate resilience, and geohazard assessment through collaborative AI systems.

🚀 Future Extensions

Possible expansions include: