Design an ETA Prediction System

Problem Formulation#

Clarifying the ML Objective#

ML framing: Given an origin, destination, departure time, and real-time traffic state, predict the trip duration in seconds as a continuous value.

This is a regression problem, but the business framing adds a constraint that makes it harder than it looks. You're predicting a future quantity (how long will this trip take?) using features that describe the present (what's traffic like right now?). The model never observes the trip it's predicting. That asymmetry between training and serving is something interviewers will probe, so get comfortable talking about it early.