What data normalization techniques do you use for JSON data, and how do you balance real‑time performance with data quality?

For JSON, I flatten nested objects into a star schema, use a schema registry to enforce versioning, and apply type casting during ingestion. To keep real‑time performance, I process the JSON in micro‑batches with Spark Structured Streaming, applying schema evolution rules to avoid full re‑processing. I also use a two‑stage approach: first write raw JSON to a raw layer, then transform to normalized form in a separate job. This decouples ingestion speed from downstream quality checks, allowing near‑real‑time loads while still ensuring data integrity.