Real-Time Fraud & AML Risk Scoring for Cross-Border Fast Payment Rails
With real-time payment linkages between Singapore PayNow and India UPI handling thousands of sub-second cross-border transfers daily, anti-money laundering (AML) detection engines must inspect streaming transaction telemetry without introducing perceptible latency to end users. In this challenge, you will construct a real-time AML risk scoring agent using LlamaIndex. The agent retrieves entity risk profiles, evaluates velocity anomalies across linked payment identifiers, and generates explainable risk evidence reports for MAS compliance officers.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a real-time AML risk scoring agent using LlamaIndex RAG over graph entity profiles and streaming transaction logs.
How work is evaluated
Evaluates AML anomaly detection accuracy and generation of compliant audit evidence.
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.
structuring_flag_check
Must flag transaction sequence as suspicious structuring.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
p99_latency_ms
99th percentile query latency in milliseconds • target: 450 • range: 0-2000
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Implement LlamaIndex index structures for real-time entity profile retrieval
Compute transactional velocity and structural graph anomalies in fast payment streams
Synthesize natural language explainable AML audit trails compliant with MAS Notice 626
Optimize agent retrieval latency to meet sub-500ms processing SLAs
Reference links and supporting material
PayNow-UPI transaction stream logs with injected smurfing, velocity anomalies, and PEP watchlist records.
Challenge setup
Evaluates AML anomaly detection accuracy and generation of compliant audit evidence.
Challenge input
JSON array of consecutive micro-transfers to linked UPI IDs within 10 minutes
LlamaIndex
Required framework for knowledge graph retrieval and transaction stream context ingestion.
Evaluated output
JSON risk score (0-100), flag_str boolean, and rationale summary
- Must flag transaction sequence as suspicious structuring.
- 99th percentile query latency in milliseconds • target: 450 • range: 0-2000
- Benchmark: FinAML-Bench
- P99 Latency Ms target: 450
- 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
- Llama Index · Required
- RAI · Optional
- Relicx · Optional
Evaluation contract
- structuring_flag_check · Weight 1
- p99_latency_ms · Weight 1
Metrics
- p99_latency_ms · Target 450
Benchmarks
- FinAML-Bench
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON array of consecutive micro-transfers to linked UPI IDs within 10 minutes
Output: JSON risk score (0-100), flag_str boolean, and rationale summary
Execution harness: Python
Evaluation tools
- LlamaIndex · RAG Framework
Recipe freeze marker
No recipe freeze time is recorded.