Real-Time BNPL Fraud Assessment Engine with Google ADK
Point-of-sale Buy Now Pay Later (BNPL) loans are highly vulnerable to synthetic identity fraud and instant account takeover. Build a real-time risk assessment agent using Google ADK and Gemini to parse applicant identity signals, evaluate device telemetry, and decision credit requests. Keep fraud loss rate below 0.1% while maintaining an instant approval latency under 300ms.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an instant BNPL credit fraud evaluation agent using Google Agent Development Kit (ADK) and Gemini to analyze applicant identity risk signals.
How work is evaluated
Evaluates BNPL loan approval accuracy, fraud prevention recall, and sub-300ms decision latency.
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.
synthetic_identity_block_test
Ensures high-risk synthetic identity attributes result in DECLINE
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
fraud_detection_recall
Recall metric for detecting synthetic fraud applications • target: 0.96 • 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
Initialize Google ADK Agent structure integrated with Gemini models
Design multi-modal input processing for applicant device fingerprints and identity records
Construct dynamic fraud risk scoring routines using Vertex AI tools
Optimize agent reasoning loops to achieve under 300ms total response time
Reference links and supporting material
Dataset of 1,000 BNPL application events with attached device telemetry and synthetic identity labels.
Challenge setup
Evaluates BNPL loan approval accuracy, fraud prevention recall, and sub-300ms decision latency.
Challenge input
JSON application payload including device_id, ssn_hash, email_domain_age_days, requested_amount
Google ADK
Google Agent Development Kit framework
Evaluated output
JSON response with decision (APPROVE, DECLINE, STEP_UP), fraud_risk_score, latency_ms
- Ensures high-risk synthetic identity attributes result in DECLINE
- Recall metric for detecting synthetic fraud applications • target: 0.96 • range: 0-1
- Benchmark: BNPLRisk-Eval
- Fraud Detection Recall target: 0.96
- 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 · Required
- Google ADK · Optional
- Google Jules · Optional
- Google Jules · Optional
Evaluation contract
- synthetic_identity_block_test · Weight 1
- fraud_detection_recall · Weight 1
Metrics
- fraud_detection_recall · Target 0.96
Benchmarks
- BNPLRisk-Eval
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON application payload including device_id, ssn_hash, email_domain_age_days, requested_amount
Output: JSON response with decision (APPROVE, DECLINE, STEP_UP), fraud_risk_score, latency_ms
Execution harness: Python
Evaluation tools
- Google ADK · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.