git-push-and-pray / reply-ai-agent-challenge-2026
-- : -- : --

Reply AI Agent Challenge 2026 — Team Entry

GIT
PUSH &
PRAY

ITIS Alessandro Rossi — Vicenza, IT
Reply AI Agent — External Edition
Multi-agent fraud detection system
LLM agents + Langfuse

Countdown to launch

--

days

--

hours

--

minutes

--

seconds

Start

16 Apr 2026 — 15:30 CEST

→

End

16 Apr 2026 — 21:30 CEST

→

Duration

6 hours

The team — 4 members

$ whoami

Alessandro
Panait

Member · ITIS Rossi

[ student ]

$ whoami

Araf
Kamal

Member · ITIS Rossi

[ student ]

$ whoami

Leonardo
Benincà

Member · ITIS Rossi

[ student ]

$ whoami

Alessia
Di Paolo

Member · ITIS Rossi

[ student ]

The
Mission

In the futuristic digital city of Reply Mirror, we are members of The Eye — a financial intelligence unit tasked with detecting Mirror Hackers: adaptive fraudsters who constantly reinvent their attack patterns.

Our challenge: build a multi-agent AI system capable of identifying fraudulent transactions in real time, even as fraud tactics evolve across 5 increasingly complex datasets.

6-hour window. Standard tooling.

Organizer Reply S.p.A.
Challenge Type AI Multi-Agent System
Datasets 5 levels (3+2 unlockable)
Team Size 4 members
Monitoring Langfuse
League High School Students' League
School ITIS Rossi, Vicenza
Scoring Accuracy + Economic Impact + Efficiency
Platform challenges.reply.com

Technical writeup

How Our
Agent Works

Post-challenge — Coming Soon
agent.py
Agent architecture — to be revealed after the challenge   $ cat architecture.md [LOCKED — results pending] ░░░░░░░░░░░░░░░░░░░░░░░   $ cat agent_roles.json [LOCKED — results pending] ░░░░░░░░░░░░░░░░░░░░░░░   $ cat results.log [LOCKED — results pending] ░░░░░░░░░░░░░░░░░░░░░░░   Documentation will be published after the race is complete Agent design, prompt engineering, scoring, lessons learned   $ git log --oneline > "still pushing, still praying" — GIT PUSH & PRAY█

01

Agent Design

Architecture and roles — post challenge

02

Prompt Strategy

LLM prompting and Langfuse traces — post challenge

03

Results & Score

Leaderboard position and analysis — post challenge

Open Source
After Results

Following the official validation of leaderboard results by Reply, we plan to release our complete agent system code publicly. Full transparency: prompts, architecture, tooling, everything.

Repository will go live once the validated leaderboard is published (within 10 working days from challenge end).

GitHub — Locked

available after results