Only one process talks to the provider
The API allows one call every 45 seconds. If the panel asked too it would compete for the quota and starve, so only the recorder talks to it and the panel reads from Postgres.
04 / 04Ride-hailing fleet ops
Luxxo
Management panel for an electric ride-hailing fleet in Barcelona: 24/7 telemetry, licences, GPS-matched fines and invoicing.
- My role
- Design, code, deployment and 24/7 operations
- Year
- 2026
Stack
Next.jsPrismaPostgreSQLSOAPLeafletDocker
- telemetry recorder
- 24/7
- provider API quota
- 1 / 45 s
- countdown before any car command
- 10 s
01The problem
An electric ride-hailing fleet working with Uber, Cabify and Bolt. Telemetry depended on the tracking provider's API, which allows one call every 45 seconds and only lets you recover the last 24 hours. They needed their own history, to know who was driving each car at any moment, and to manage licences, fines and invoices in one place.
02What I built
- 0124/7 daemon over the SOAP tracking API, with an in-house history that can be recomputed
- 02Live map, trips, utilisation and driving style per vehicle
- 03VTC licences and driver assignments with their validity
- 04Fines matched by GPS to whoever was driving the car at the time
- 05In-house invoicing and PDF/CSV reports
- 06AI assistant on a read-only database role
03Decisions that matter
The history is ours
No history screen or report depends on the provider answering. Trips and daily stats are derived from stored positions and can be recomputed; positions are never deleted.
The cars are real
Any command that acts on a vehicle is superadmin-only, goes through a 10-second countdown and lands in the audit log. The AI assistant has no tool that can touch them.
Money in cents, days in Madrid
Amounts are always integer cents and dates use Temporal. Every per-day aggregate is cut in Europe/Madrid, never in the server's time zone.
04In motion
VTC 4821 · Tesla Model 3
km/h32batería78%señalSOAP ● live
Uber …
Cabify …
Bolt …