The starting point
The records lived in WhatsApp chats.
After every event, NightGuides documents what happened on the floor. Every incident, every intervention, every code raised during a shift goes into a report.
That record started life inside WhatsApp group chats. Hundreds of messages per event: severity codes, guest counts, and case updates, mixed in with coordination, logistics, and noise.
Turning that into a clean, client-ready report meant a staff member copying the chat into a document by hand and shaping it into something readable. It took about four days. Every meeting that depended on those reports had to wait for them.
What I built
One form. Two-stage AI. Private server.
The team finishes their shift and submits the night's WhatsApp export and the event briefing through one short form. From there it runs on its own.
A first pass reads the raw chat and pulls out what actually matters: the incidents, the codes, the timeline, the guest counts, who was on, how each case was handled. It cuts the side-conversation, the coordination, and the noise.
A second pass writes the report in the format the team already reviews on: what happened, where, when, who was involved, how it was resolved, with the serious cases written up in full.
The finished report is a formatted Google Doc, ready within minutes of the submission. No manual typing. No copy-paste. No waiting on a manager to clean it up.
Two things make it work: a two-stage AI chain that separates reading from writing, so nothing gets invented and nothing gets lost, and a private server that runs the whole pipeline, so the incident data stays on infrastructure NightGuides controls.
