AI Agent Challenge 2026 - Final Archive

This directory is prepared for GitHub publication as a completed challenge archive.

Status: challenge finished.

Final Result

• Leaderboard position: 138 / 1971 teams

What This Repository Contains

This project stores:

• official challenge material and rules;
• train and validation datasets used during the competition;
• a shared agentic fraud-detection script;
• produced validation outputs.

Agentic System Design: The solution implements a multi-agent ReAct orchestrator with specialized components:

• Data Analyst Agent - Extracts pattern signatures from transaction history, user personas, and behavioral context
• Anomaly Detection Engine - Applies heuristic baseline + LLM-based decision making with economic impact awareness
• Review Agent - Secondary review pass for uncertain/disputed cases to improve precision-recall balance

Key Optimizations:

Metric	Initial	Optimized	Outcome
Recursion Limit	90	70 (adaptive by dataset)	30% token overhead reduction
Token Usage	1800 (max)	1200 (max)	Cost efficiency maintained
Fallback Models	3 duplicates	1 unique	Failure cascade prevention
Z-Score Threshold	3.0	3.5	Enhanced outlier detection
Model Invocations	Full batch	Adaptive calibration	Budget-aware selection

Detection Quality Improvements:

• Enhanced system prompt with explicit fraud signals (behavioral anomalies, economic misalignment, channel anomalies, location contradictions)
• Implemented balance-impact scorer detecting transactions causing >50% balance drops
• Added whitelisted legitimate patterns (salary, recurring utilities, subscriptions) for false positive reduction
• Calibrated ranking to prioritize high-value fraud detection

Technology Stack:

• LangChain (agentic orchestration), OpenRouter API (LLM access), Langfuse (observability & tracing)
• Environment: Python 3.10+, Jupyter for experimentation, Makefile for reproducible setup
• Infrastructure: Full .env-based credential management, token budget tracking, submission session logging

Directory Structure

02_AI_Agents_Challenge/
├── README.md
├── 00_How_It_Works/
│   ├── README.md
│   ├── api_guidelines.md
│   ├── model_whitelist.md
│   ├── submission_guide.md
│   └── challenge_day_checklist.md
└── 01_Material_Implementaion/
    ├── AIAgentChallenge-ProblemStatement16April.md
    ├── 01_The+Truman+Show+-+train/
    ├── 02_Brave+New+World+-+train/
    ├── 03_Deus+Ex+-+train/
    ├── Script/
    │   ├── main.py
    │   ├── agent.py
    │   ├── tools.py
    │   ├── requirements.txt
    │   ├── .env.example
    │   └── README.md
    └── validation_outputs/

Notes:

• __MACOSX directories are extraction artifacts and can be ignored.
• Folder name Implementaion is intentionally preserved to avoid breaking existing paths.

Dataset Compatibility With Shared Script

The shared script in 01_Material_Implementaion/Script/ is reusable across datasets that expose the expected schema.

Required files:

• transactions.csv
• users.json
• locations.json

Optional files:

• sms.json
• mails.json

Current availability:

Dataset block	Train files present	Validation files present	Runnable with Script
The Truman Show	Yes	Yes	Yes
Brave New World	No	Yes	Validation only
Deus Ex	Yes	Yes	Yes

Quick Start

1. Move into the shared script folder:

cd 01_Material_Implementaion/Script

2. Create and activate a Python virtual environment.

3. Install dependencies:

pip install -r requirements.txt

4. Create your environment file:

cp .env.example .env

5. Fill .env with your keys and team name.

6. Run on a dataset folder that contains the required files:

python main.py \
  --dataset "../01_The+Truman+Show+-+train/The+Truman+Show+-+validation/The Truman Show - validation" \
  --model "deepseek/deepseek-v3.2" \
  -o "../validation_outputs/output_truman_validation.txt"

Publish Readiness Notes

• .env is ignored through .gitignore; only .env.example is committed.
• Python cache/editor temporary files are ignored.
• Existing output files are kept as experiment artifacts.