English · Deutsch
Scope Dispatch is a local planning application for a fictional technical service team. It matches jobs to qualified people, checks arrival windows and travel times, and lets a dispatcher review changes before accepting them.
I developed it with Codex to explore a different part of industrial software: organizing the people who maintain cameras, robots and production equipment.
The team schedule takes the main space. Arrivals, journeys and breaks sit on one timeline; the route atlas and a light order desk provide the detail below. I chose this layout so a dispatcher can follow the day before looking at an individual job.
A camera inspection system needs urgent attention. Scope Dispatch adds the job, calculates a new schedule and shows which appointments would change. A fixed appointment keeps its technician and arrival time. When a team member is absent, the remaining work can be redistributed.
Watch the screen recording · Planning decisions and implementation
The map and timeline use the same calculated plan. Routes follow a fictional road network with fixed travel times. They are not live navigation or traffic data.
- Plan a service day. Five technicians, four skill areas, arrival windows, job durations and a shared lunch break at the service base.
- Review changes. Every proposal shows its effects before it is saved. Unassigned work remains visible.
- Protect commitments. Lock an appointment to its technician and exact arrival time.
- Reassign a job. Drag it onto another technician's row, or use the selector. The application checks qualifications and availability before proposing new times.
- Prepare a request with local AI. A language model extracts a draft from a service message. Missing or unsupported fields remain empty. A person reviews the details and chooses the customer and location.
- Keep the confirmed plan accessible. A compact technician view, a change history and a local export all use the accepted state.
The recorded normal-day comparison scheduled all 20 jobs with both methods. Total travel was 640 minutes with arrival-order insertion and 468 minutes with the bounded search. In the combined urgent-job and absence scenario, the search scheduled 21 of 21 jobs, compared with 19 of 21 for the baseline.
These are measurements from the fictional dataset. They are not customer savings or a claim that the schedule is globally optimal. Search results can vary with available processing time.
37 automated tests passed. Six local-model examples also matched all 24 expected structured fields after validation. That small regression set does not establish general language-model accuracy. The verification notes include the comparison table, browser checks, model mistakes found during development and remaining limitations.
This repository contains documentation and recordings. Application source, executable files, configuration, database, model files and raw test artifacts remain private. There is no publicly hosted application.
Original work is not offered under an open-source or Creative Commons license. Rights · Third-party software.
Independent project by LeonScope, developed with Codex.