Skip to main content
Back to challenges
Challenge

Scalable UPI Merchant Fraud Triage Agent using Google ADK and Galileo

With UPI transactions hitting 23.7 billion monthly, the volume of potential merchant-side fraud has scaled exponentially. This challenge requires building an automated fraud triage agent that processes high-velocity transaction logs and merchant metadata to identify high-risk patterns. You will leverage the Google Agent Development Kit (ADK) to build a multimodal agent that uses Gemini 1.5 Pro to analyze both structured transaction data and unstructured merchant verification documents (KYC images, storefront photos). The agent must utilize Galileo's monitoring suite to implement real-time guardrails, ensuring that automated fraud flags meet high precision standards before triggering account freezes. Your solution will need to handle India-specific context, such as diverse merchant categories and regional transaction behaviors, while maintaining the performance required for national-scale Digital Public Infrastructure.

Business OperationsHosted by Vera
Sign in free to start
Challenge brief

What you are building

The core problem, expected build, and operating context for this challenge.

Build an AI agent using Google ADK and Galileo to automate UPI merchant risk assessment and fraud detection at scale.

Delivery guide

How work is evaluated

Evaluation

The system is evaluated based on its ability to correctly identify fraud in a synthetic UPI dataset and the latency of its multimodal analysis.

Datasets

Challenge datasets

Public datasets, plus the private datasets you have access to.

Evaluation rubric

How submissions are scored

These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

High Risk Flagging

Must flag accounts with >100% volume increase in 1 hour as high risk.

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

Precision

Percentage of correctly identified fraud cases. • target: 0.9 • range: 0-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

What you should walk away with

  • Master high-velocity data ingestion and analysis using Google ADK and Gemini

  • Implement automated fraud triage logic for UPI merchant ecosystems

  • Integrate Galileo for LLM observability and automated quality guardrails

  • Design multimodal workflows that combine image analysis with tabular transaction data

Resources and assets

Reference links and supporting material

Dataset notes

Synthetic UPI transaction logs representing 10,000 merchants with associated KYC images and mock risk labels.

Challenge setup

The system is evaluated based on its ability to correctly identify fraud in a synthetic UPI dataset and the latency of its multimodal analysis.

Challenge input

JSON containing merchant_id, transaction_history (list), and kyc_image_url

Google ADK

Required for building the multimodal agent logic.

Galileo

Ensures reliability and guardrails for automated risk decisions.

Gemini 1.5 Pro

Used for its large context window and multimodal capabilities.

Evaluated output

JSON containing risk_score (0-1), reasoning_string, and recommended_action

Evaluation checks
  • Must flag accounts with >100% volume increase in 1 hour as high risk.
  • Percentage of correctly identified fraud cases. • target: 0.9 • range: 0-1
Configured evaluation evidence
  • Benchmark: FraudDetection-v1
  • Precision target: 0.9
  • 1 public reference case
Configured runtime
  • 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

Action Space
  • Google · Optional
  • Google ADK · Optional
Observation
  • Galileo · Required
Reward / Eval
  • Galileo · Required

Evaluation contract

  • High Risk Flagging · Weight 1
  • Precision · Weight 1

Metrics

  • Precision · Target 0.9

Benchmarks

  • FraudDetection-v1
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON containing merchant_id, transaction_history (list), and kyc_image_url

    Output: JSON containing risk_score (0-1), reasoning_string, and recommended_action

    Execution harness: Python

Evaluation tools

  • Google ADK · Orchestration Framework
  • Galileo · Evaluation & Monitoring
  • Gemini 1.5 Pro · Model

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Scalable UPI Merchant Fraud Triage Agent using Google ADK and Galileo