_{Edited By PhotoFilter}

Hello steemians,

The email hit at 13:01 in the Lac 1 office. “Meeting room after lunch. Priority build.” At 15:00 the blinds rattled in the lake breeze while John stacked a notebook over the trembling coffee cups. Timothy looked like he had been expecting the summons for weeks.

The boss spoke without slides. “A car rental client wants dynamic pricing. Real time. Peak and off-peak awareness. Promotions that do not eat revenue but keep fares competitive. Six weeks. John and Timothy, assemble your engineers today and give me a first approach before you leave.”

Out in the corridor, John said, “We are already deep in the payment refactor.”

Timothy answered, “In Tunis you learn to merge in traffic. We will not stall in the roundabout.”

They took a war room and wrote a short creed on the whiteboard: Fair. Competitive. Sustainable. The team list followed in quick strokes. Aïcha for data science. Marco for backend. Lila for Android and iOS. Wei for SRE. Sofia for UX. Yassine, a Tunisian intern with a sharp eye for messy data, would assist Aïcha.

Plan. Week 1, collect signals that actually move demand: historical bookings, Tunis traffic feeds on Avenue Habib Bourguiba and the Ring Road, weather from Carthage Airport, ferry and flight arrivals, the Carthage summer concerts, weekend travel toward Sousse and Hammamet. Week 2, model demand with explainable features. Week 3, design the pricing engine with constraints. Week 4, promotions tied to elasticity and fleet balance. Week 5, experiments in a sandbox with canary release. Week 6, hardening, docs, handover.

“Ethics belongs in the spec,” Timothy said. “Publish a surge cap. Explain every price in clear French and Arabic. No gouging on rainy nights.”

Execution started with noise. At 03:07 every Tuesday, bookings near Tunis Marine spiked for five minutes, then vanished. “Either a ghost fleet sails in,” Yassine said, “or we have bots.” He traced the pattern to a legacy kiosk that sent test reservations each night when the SNCFT schedule updated. John drew a skull next to the pipeline labeled Null. They walled the stream and rebuilt their feature store so dirty events could not leak into training or live pricing.

Aïcha trained a hybrid model. Gradient boosting for short term lift. Seasonality priors for rhythms like Friday exodus to the Sahel. Weather and event modifiers for rain, heat, and festival nights in Carthage. Every prediction shipped with a small explanation vector that Sofia could translate into human words inside a drawer titled Why this price.

John framed the price engine as an optimization with guardrails. Maximize contribution margin subject to a surge cap of plus 20 percent, a competitor parity band, a minimum acceptance rate, and geographic fairness so under-served neighborhoods like Ettadhamen were not penalized by old bias. The solver had to answer inside 60 milliseconds. Marco tuned caches until it did.

Promotions became strategy. “We do not spray discounts,” Timothy said. “We move supply like caravans.” Lila grinned. “Then call it Promo Caravan.” When La Marsa ran dry, the app could offer riders a small discount to go there only if there was a strong chance that the car would catch a profitable return from the beachfront later. The design framed it as a civic favor. Help rebalance. Save now.

Midway, the stand-up turned into a huddle.
“Tonight there is a major concert in the amphitheatre,” Aïcha said. “Rain likely at 20:00. Demand will jump and stick.”
“Hold the surge cap,” Timothy said.
“That costs margin,” John replied.
“We take it back with Promo Caravan after the encore,” Aïcha said. “And a tiny loyalty perk to keep reliable drivers online for twenty minutes.”
Wei counted with his fingers. “Canary at five percent. Rollback if p95 latency crosses 200 ms. Drift alerts on feature distributions. Backup nowcasting if the event bus coughs.”
Marco nodded. “Feature store read through. If a feature is late, backfill with the last good snapshot for two minutes.”

The night before the demo, the event stream failed. A cable cut in the data center. Wei switched to degrade mode. Persisted features. A featherweight nowcaster. Parity bands from the cache. Prices remained coherent. Explanations did not lie. Yassine wrote in the incident doc that reliability is also a product feature.

Result day arrived with sun on the lake and heat shimmering over Route X. The boss and the client watched a simulated Saturday. At 20:00, demand rose by 33 percent during a thunderstorm. Prices stayed under the cap. Acceptance fell by four points then recovered as loyalty perks held the fleet online. Promo Caravan moved 96 cars toward blue zones and cut outer neighborhood pickup times by 15 percent. Net revenue improved by eight percent. Cancellations dropped because Sofia added a ten minute Price Hold that stabilized nerves.

“Explain one fare,” the client said. Sofia opened the drawer. Rain plus stadium event added eleven percent. Local supply added five. Loyalty subtracted three. Competitor parity subtracted two. Surge cap enforced. Final adjustment plus eleven.

“What if there is another shock,” the boss asked, “like borders closing or curfews.”

John said the engine would flip to a client approved ruleset with public thresholds if the inputs left the training manifold. Timothy added that they would show the exact rule in the explanation. No black boxes on a bad night.

Approval felt like a breeze from the lake. Outside, the real city kept moving. Timothy’s phone buzzed. Promo Caravan to Sidi Bou Said. Help rebalance. Save 12 percent. He tapped Accept and watched the car icon glide up the coast, a small dot in a system designed to be fair, competitive, and sustainable.

---

Thank you very much for reading, it's time to invite my friends @sualeha, @drhira, @shiftitamanna to participate in this contest.

Best Regards,
@kouba01