Resumable Fact Table Load

You work as a data engineer for a retail analytics company. You are tasked with designing a data pipeline that loads transactional data into a fact table. The requirement is that if the loading process fails partway through (for example, after successfully loading 100 out of 250 records), a rerun should resume loading from the point of failure, ensuring no duplicate records are loaded and none are missing.

How would you architect this loading process?