landinc

ai engineer · apr 2026 – now

a platform that reads ten procurement portals every morning so nobody at the firm has to.

landinc bids on public work, and finding the right rfp used to mean someone opening ten portals, searching each one by hand, and skimming pdfs to decide what deserved a second look. i lead the engineering on the platform that replaced that routine: it watches every portal, reads each new posting, and puts a short, ranked list in front of the team.

python · playwright · postgres · llm scoring

problem: every portal has its own search, its own session rules, and its own idea of what a closing date looks like. review was slow, inconsistent between people, and easy to skip on a busy week, which is exactly when good opportunities slipped past.
collection: one watcher per portal, each a small playwright job with its own selectors and login flow, writing raw postings to one table. when a portal changes its markup, that watcher fails loudly and the other nine keep running, so one broken site never means a blind morning.
scoring: new postings are normalized, deduplicated across portals, and scored against what the firm actually wins: discipline, region, size, and timeline. a model reads the attached documents and writes a two-line case for or against, so a reviewer can agree or disagree in seconds instead of reading forty pages.
review loop: the team gets one ranked digest, about 20 qualified opportunities a week, with the reasoning attached and one click to keep or dismiss. every dismissal is a label, and the scoring learns from them.
outcome: about ten hours of manual review saved every week, and a pipeline the team trusts enough to stop checking the portals themselves.

updated oct 2026next