The problem
A factory needs 400 units of a part. Someone emails eight suppliers, chases them for three days, then copies the replies into a spreadsheet, each in a different format with different terms and units. That is a full day of skilled time for one purchase, every week.
What we built
An agent that sends the RFQ to the supplier list, reads the replies in whatever format they arrive, extracts price, quantity, lead time and terms, converts them to a common basis and puts them in one comparison table. It follows up with suppliers who haven't replied and flags anything uncertain for a person to review.
How it was built
Sourcora is built in Python, using language-model extraction with a confidence threshold, PostgreSQL and email integration.
Why it matters
This is what a useful AI agent looks like: one specific, time-consuming task handled end to end, with a person checking the parts that matter.
