A title research company · Outside technology lead · 2025 to 2026
Starting the automation one step earlier
For a title research team that sorted every search by hand before software could read it
Built with Patrick Hrabos
The result
No more sorting by hand
Once the new system went into production, the company’s staff stopped bookmarking title files by hand.
The short version
The problem
A person sorts it first
Each title search came in as a single long file, and before the extraction software could start, someone on the company’s team had to group and bookmark its sections by hand.
What we built
Reads the whole file
Jeremy Brooks, Signal Scout’s founder, made skipping that step a requirement, and we built a system that reads every page and does the bookmarking itself before pulling out the data for review.
Jeremy’s role
With the experts and the builders
He sat with the company’s experts and turned what they needed into the requirements and the architecture, and Patrick Hrabos led the engineering. Jeremy also designed the second phase, where rules check the data before a person sees it, and we handed it to the company’s own team.
The step everybody had stopped seeing
A title search arrives as one long PDF that mixes deeds, mortgages, liens, maps and tax records of every quality, from clean print to handwriting and poor scans. The title research company we worked with already had software to pull the data out of those files, but before it could start, someone on their team had to go through each file by hand, grouping its sections and bookmarking them, and that was simply how every search began.
That step added work for their team on every single search, and any automation that started after it would leave that work right where it was. So Jeremy made it a requirement for the new system to skip manual bookmarking from day one and work out on its own what each page is and where it belongs.
Sitting with the people who knew the work
Jeremy worked with the people who knew title work and the people building the software. With the company’s leaders and its title and quality experts, he learned what they needed from a file, and he turned that into the requirements and the architecture we built from.
The work started from a proof of concept Jeremy had built, and Patrick Hrabos led the engineering from there and built most of the first phase.
We built it as one general system rather than one tied to a single state, and it went into production. The screen we first made to check our own work became the one the company’s staff worked in once the system was live.
Letting rules make the call
The company also wanted to automate the examination and quality review that follow extraction, and title examination is a judgment with legal weight. Jeremy designed the second phase so that AI would read the documents and turn them into data, and every examination decision would come from an explicit rule that a person could read and audit.
The experts carried knowledge that lived only in their heads, so we met with them daily and turned it into checks one at a time. In that design, people see only the cases that fail a check, so the experts review the exceptions and leave the rest to the rules.
We proved the design on the first of those checks and handed it to the company’s own team.
How the two phases were designed to fit together
- A title search arrives as one file
- Each page is identified and the sections bookmarked
- The data is pulled out for people to review
- Explicit rules check that data
- People see only what fails
Getting what a company’s experts know into what its engineers build is the work we do with CEOs and their teams now. See how the three steps work.
