Design a scalable job to compute rolling 15-minute meeting concurrency and room utilization

You are given a massive log of meeting events with fields:

• user_id
• start_time (UTC)
• end_time (UTC)
• room_id
• office_id

Compute, for each rolling 15-minute window (assume slide = 1 minute unless specified otherwise):

1. Concurrent meeting counts per office (report peak concurrent meetings within each 15-minute window).
2. Rooms whose utilization exceeds 80% within the window, where utilization = occupied_time_in_window / 900 seconds.

Design a MapReduce or Spark job that:

• Specifies map keys/values and how you transform intervals into windowed aggregates (e.g., line-sweep with +1/−1 at boundaries vs. interval splitting into windows).
• Describes the shuffle/partition strategy to minimize skew and bound reducer work.
• Explains handling of late and duplicated events (event-time watermarks, dedup keys, update/retraction strategy).
• Proposes validation and correctness checks at scale.

Make minimal, explicit assumptions if needed (e.g., presence of meeting_id for dedup, window slide granularity).