Public-notice processing

The problem

The client asked a plain question: “Can you get me foreclosure notices when they hit the 20-day public-notice deadline? I only want pre-foreclosure leads that are 20 days from auction.” No feed existed for that. Someone had to find where the notices were posted, work out which ones qualified and get them to the client with enough property detail to act on, the same day.

What I built

I built and operated a pipeline that found the requested notices, added property details, removed duplicates and emailed qualified records.

The result

Delivered same-day alerts and 30+ qualified leads per month from the specific notice stream the client requested, supporting daily follow-up.

How it fit together

I found the source and chose the stream: the public site where the notices were posted, and the 20-day window the client cared about. A Python pipeline then read each notice as unstructured text, extracted fields into Pydantic schemas with an LLM, added property details, removed duplicates, applied qualification rules and offer ranges, held uncertain or incomplete records for a person, and emailed the qualified ones the same day. The novel part was the first step; nobody had packaged that stream before.

Pydantic schemas defined the extracted fields. Persistence and failure states kept incomplete records visible for review.

Decisions and tradeoffs

The model extracts; rules qualify.

Explicit rules need maintaining by hand, but every qualification and offer range can be explained to the client in one sentence.

Keep incomplete records visible instead of dropping them.

More review work for the operator; nothing is silently lost when a notice is missing a field.

AI extracted information from unstructured text. Explicit rules handled qualification and offer ranges; uncertain records went to a person for review.