A data engineering team is designing a storage strategy for a repository that will ingest large volumes of raw sensor logs, social media feeds, and image files. Analysts want the flexibility to define the structure of the data only at the time they query or process it, rather than being forced to conform incoming data to a predefined table structure during ingestion. Which approach best matches this requirement?