S07 · interview simulation

System design: a demand forecasting platform for 2,000 route-days

People, data and outputs in this scenario are simulated. Each stage is recorded when you commit; the reveal comes after.

5stages
45 minabout
0stages committed

The brief

Simulated interviewer: "Our airline (a simulated 22-route network, larger than the course database's 11 routes) wants a platform that forecasts demand for every route and departure date in the next 91 days, about 2,000 route-days (22 x 91 = 2,002), refreshed nightly. Revenue management uses it for fare-class availability and overbooking. Design it." History: three years of daily data with booking-curve snapshots at 13 days-to-departure points. All facts in this scenario are simulated.

Stage 1 of 5

You have five minutes to ask questions before designing.

Which questions shape the design most? Select all that apply.

Which decisions consume the forecast, at which days-to-departure, and how often are they re-run?Is the target bookings as sold, or unconstrained demand including requests turned away?Do consumers need intervals or quantiles, and at which levels?How many GPUs can we use for training and serving the forecasting model?Should we use a transformer, given how well they perform on sequence forecasting benchmarks?