Many real-world datasets start as unstructured HTML. I used Python and BeautifulSoup to crawl paginated listings, extract fields with defensive selectors, and normalize records into JSON/CSV. Edge cases matter: missing fields, duplicate entries, rate limits, and encoding issues. I added retries, delays, and validation before export. For data/AI roles, this is foundational — ingestion and normalization are always the first step before any model or search index can work reliably.