Anti-Scam Root Cause Triage Agent with Google ADK and Gemini
Scam operations target citizens using evolving phishing mechanisms, demanding rapid intelligence extraction and coordinated public safety response across agencies. Build a Google ADK agent utilizing Gemini multimodal reasoning to ingest scam SMS reports, correlate suspect URL clusters, and generate actionable anti-scam intelligence bulletins.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop an automated anti-scam triage pipeline using Google Agent Development Kit (ADK) and Gemini multimodal models to process public scam reports.
How work is evaluated
Evaluates Google ADK agent accuracy in identifying scam domains and outputting standardized incident response directives.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
takedown_trigger_test
Ensures domain takedown recommendation is issued for active Singpass phishing sites
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
extraction_precision
Accuracy of extracted phishing URLs • target: 1 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Build modular AI agents using Google Agent Development Kit (ADK) and Vertex AI
Analyze multimodal scam evidence (screenshots, SMS text, phishing domain records)
Cluster scam infrastructure patterns using similarity search and graph linking
Automate public warning bulletin generation and ISP domain takedown requests
Reference links and supporting material
Anonymized phishing SMS feeds and domain registry logs from Singapore scam reporting channels.
Challenge setup
Evaluates Google ADK agent accuracy in identifying scam domains and outputting standardized incident response directives.
Challenge input
JSON containing SMS text body and base64 encoded screenshot metadata
Google ADK
Google Agent Development Kit providing enterprise integration with Vertex AI and Gemini
Evaluated output
JSON containing extracted_urls, impersonated_entity, threat_level, and takedown_recommended
- Ensures domain takedown recommendation is issued for active Singpass phishing sites
- Accuracy of extracted phishing URLs • target: 1 • range: 0-1
- Benchmark: Cyber Threat Triage Benchmark
- Extraction Precision target: 1
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- Google ADK · Required
- Google Gemini · Optional
- Google · Optional
Evaluation contract
- takedown_trigger_test · Weight 1
- extraction_precision · Weight 1
Metrics
- extraction_precision · Target 1
Benchmarks
- Cyber Threat Triage Benchmark
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing SMS text body and base64 encoded screenshot metadata
Output: JSON containing extracted_urls, impersonated_entity, threat_level, and takedown_recommended
Execution harness: Python
Evaluation tools
- Google ADK · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.