The problem, for readers outside Peru
In Lima and Callao, most public transport is not run by the state. It is run by hundreds of private companies, each holding one fixed route and operating dozens of microbuses. Each company is a small organization: an office, a depot, a manager and dozens of drivers and fare collectors.
Criminal groups charge these companies a "cupo": protection money, usually a fixed amount per bus per day. The demand arrives as WhatsApp voice notes and messages from disposable phone numbers, as photos of the bus or the driver's family, as letters left at the depot, and as phone calls. Those who refuse get shots fired at their buses and depots, and drivers have been murdered in front of their passengers. Transport workers have stopped the city several times in protest.
The scale, from official statistics and published interviews:
- 26,585 extortion complaints in Peru in 2025, up from 22,361 in 2024 (INEI with National Police data).
- More than 85,000 complaints produced only 20 sentences in five years (Caretas).
- About 400 urban transport companies in Lima and Callao, about 80% of them subject to extortion, many by several groups at once; the longer the route, the more groups (Martin Ojeda, general manager of the interprovincial transport association, broadcast interview).
- "Almost 100% of transport operators are afraid... extorted not by one gang, but by two or three" (Hector Vargas, president of the Coordinadora de Empresas de Transporte Urbano, on Exitosa; translated).
Why reporting falls short today: the evidence is scattered across many personal phones, drivers delete it out of fear, the phone numbers change constantly so every complaint looks isolated, and what reaches the police is a pile of loose files with no order and no proof of integrity.
I did not interview drivers for Round 1; the problem description relies on these public sources.
Inspiration
A company cannot stop the gangs, but it can control one thing: the quality of its own evidence. A bus company already has a place where everyone meets every day, the depot. A NAS there, with a private folder per driver and local AI, can do what no cloud chatbot should be trusted with: keep victims' evidence in-house, verify it, and connect the cases. This assumes the company or its association keeps a UGREEN AI NAS at the depot; that is an adoption assumption, not something I have observed.
What it does
Extortion Case Hub is an AI Skill for UGREEN AI NAS. Drivers' phones back up their chat media to a private folder on the company NAS. On request, the Skill reads every voice note, screenshot, exported chat, photo and payment capture, fingerprints each original with SHA-256 and produces:
- a driver report ready for the police complaint;
- a company report with cross-member matches (same collection account, identical file, same alias or demand pattern), an evidence-based escalation level per member, cupos paid and incidents per route segment;
- an anonymized indicator file that another transport company can cross-check on its own NAS.
It reports verifiable coincidences, never conclusions. It does not identify suspects, contact anyone or advise paying, and nothing leaves the NAS.
How I built it
With the official UGREEN Skill Developer Package (Uliya Skill Kit 0.7.7). The package is a single SKILL.md plus seven Markdown references. The Skill only delegates to registered NAS experts: file-agent, photo-agent, knowledge-agent, document-agent and, optionally, security-monitor-agent. All read steps run before any write, and every write waits for platform approval.
Challenges
Designing matches around what extortioners cannot easily change (collection accounts, identical files, aliases) instead of disposable phone numbers, and wording every result so it states a fact without accusing anyone.
Accomplishments
The package passes the official static checker with 0 errors and 0 warnings, and it includes a complete synthetic example and a validation plan with known expected results.
What I learned
On a NAS, the value is not one clever prompt. It is volume, privacy, multi-user folders and evidence integrity working together.
What's next
If selected as a finalist: run the planted-answer test set on the assigned UGREEN AI NAS, measure Spanish transcription quality, and record the demo.
Built With
- ugos
- ugreen-ai-nas
- uliya-skill-kit