Loan Match Uses One AI Call to Match 10,000 People With Personal Loans

A developer built Loan Match, a system that takes a CSV of people, reads lender websites daily, and emails each person a tailored list of loans they may qualify for, with reasons. The design began from fixed requirements rather than tool choice, and the file bypasses the API entirely: the browser uploads directly to S3 in checksummed chunks via presigned URLs, since API Gateway's size and timeout limits would fail on large files. A queue sits between S3 and the Lambda processor, providing visible retries, a dead-letter queue for failures, and a concurrency cap to protect the database. The Lambda streams rows in constant memory, normalizes messy spreadsheet data, and bulk-loads valid records into PostgreSQL, rejecting the whole import if more than 20% of rows are invalid. The author reports that the Go rewrite loads one million rows in about six seconds.