>Prompt|Operations Hackathon
HomeScheduleChallengesTeamsProjectsToolsInfo
Sign in
HomeScheduleChallengesProjectsMy team
>Prompt
An Aaltoes initiative

Bringing the AI frontline to Finland — a movement of builders, founders and experts, operated by agents.

Prompt
prompt.aaltoes.com →
Community
Telegram chat →Luma calendar →
Partner
promptagents@aaltoes.com
© 2026 PROMPT · AN AALTOES INITIATIVE · ESPOO
← All projects
OPERATIONS HACKATHON · OCT 2026

Perille

A control room for Bahn's operations team, and one agent that makes the pickup calls.

## The problem

Bahn delivers 25,000 used cars a year across Finland, Denmark, and Germany, on a promise of two days. About fifty people in operations run those deliveries, and they depend on people who do not work for Bahn: the seller holding the car, the driver, the ferry line, and whoever signs at the other end. Any of them can be late, wrong, or silent.

None of those cases is hard on its own. The hard part is that each one starts a chain, and today a person in operations holds the whole chain in their head.

## What we built

**The control room.** Only the orders that need a person, each carrying the reason it surfaced, everything else out of the way. Open one and the chain is on a single screen: what happened, who said what and when, what the agent already found out, and the one next decision. Every proposal carries the IDs it relied on, so a person can check the agent's reasoning instead of rebuilding it from the phone log.

**The agent: pickup readiness.** It reads the documents attached to an order, then calls the seller or pickup place to confirm the car and papers are actually ready before a driver is dispatched. It does the cheap work first, and when it hits something it cannot settle, it stops and hands a person what it found plus a recommended decision. Asking and looking up need no permission. Changing a booking, a date, a cost, or messaging a customer waits for a human yes.

## Where it stands

Running against Bahn's synthetic dataset: 19 alerts, 12 open, 12 proposals generated, all 12 carrying evidence references. The agent has a deterministic path that runs with no model key, and a model path that engages when a key is present, and the fallback records why it fell back.

The code being judged is tagged v1.0-hackathon at commit fbcf774 on the public repository, so the commit named here is the commit that runs.

Built during the Prompt Operations Hackathon, 24 to 25 October 2026. This page is the live submission and is edited until the deadline: the demo video, screenshots, and the public repository with run instructions land here. Anything not working is marked as not working.

GitHub repo ↗
Built by · Kia.M
  • KMKia.M
ChallengeBahnThe Car Delivery Control Room

More from this hackathon

See all →
key high density dashboard: what if VelhoVelho AI IntelligenceFind automation opportunities and turn company research into a personalized sales pitch.Sell me this hackathonBahnVelhoThis is the home page of the agentJob-ad-X-rayJob-adX-ray uses industry benchmarking and executive profiling to deliver data-proven, hyper-personalized C-suite outreach.PlannerzzVelhoLanding page with highest priority task
★ Honourable mention · Bahn
Bahn DeskA dispatch-risk control room that scores every car delivery and shows a dispatcher only the orders that need a person.aBARofchocolateBahnVelho