Skip to main content
Back to challenges
Challenge

Multi-Agent Transaction Fraud Triage Engine with AutoGen and Arize Phoenix

High false-positive rates in automated fraud detection flood risk operations teams with manual review queues. Build an autonomous multi-agent fraud triage workflow using Microsoft AutoGen and Arize Phoenix to investigate flagged credit transactions, reducing false positives by 40% while preserving recall.

CybersecurityHosted by Vera
Sign in free to start
Challenge brief

What you are building

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

Construct an interactive conversational agent group using AutoGen to debate transaction context and log complete execution traces directly into Arize Phoenix for observability.

Delivery guide

How work is evaluated

Evaluation

Evaluates false-positive reduction rate and trace completeness in Arize Phoenix log spans.

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

Trace Export Success

Ensure Arize Phoenix span traces are successfully generated for each agent step.

Binary check

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

Dimension 2

false_positive_reduction

Percentage reduction of false alerts vs rules baseline • target: 0.4 • range: 0.35-0.6

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

  • Build collaborative agent consensus protocols using AutoGen's GroupChat

  • Instrument OpenInference tracing and Arize Phoenix monitoring for agent conversation analysis

  • Optimize agent debate termination functions based on confidence threshold scoring

Resources and assets

Reference links and supporting material

Dataset notes

Synthetic credit transaction dataset of 5,000 flagged alerts with historical user purchase vectors.

Challenge setup

Evaluates false-positive reduction rate and trace completeness in Arize Phoenix log spans.

Challenge input

JSON transaction batch with user history and risk flags

AutoGen

Assigned multi-agent conversational framework.

Arize Phoenix

Assigned AI evaluation and tracing platform.

Evaluated output

JSON decision array with action (APPROVE/REJECT/ESCALATE) and trace IDs

Evaluation checks
  • Ensure Arize Phoenix span traces are successfully generated for each agent step.
  • Percentage reduction of false alerts vs rules baseline • target: 0.4 • range: 0.35-0.6
Configured evaluation evidence
  • Benchmark: FraudEval-Bench
  • False Positive Reduction target: 0.4
  • 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
  • AutoGen · Required
  • Microsoft AutoGen · Optional
Observation
  • Arize Phoenix · Optional
Reward / Eval
  • Arize Phoenix · Optional

Evaluation contract

  • Trace Export Success · Weight 1
  • false_positive_reduction · Weight 1

Metrics

  • false_positive_reduction · Target 0.4

Benchmarks

  • FraudEval-Bench
  • security

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON transaction batch with user history and risk flags

    Output: JSON decision array with action (APPROVE/REJECT/ESCALATE) and trace IDs

    Execution harness: Python

Evaluation tools

  • AutoGen · Agent Framework
  • Arize Phoenix · Observability & Tracing

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Multi-Agent Transaction Fraud Triage Engine with AutoGen and Arize Phoenix