Leonida Desk

Leonida Desk is a test of how much of a digital media company AI automation can run end to end. Automated news fails in familiar ways: duplicate coverage, invented sources, overstated rumours and low-quality rewrites. The question was whether a system could find, write and publish gaming news to a live site while one human editor stays in control.

I designed a pipeline that finds, ranks, researches, writes, checks and publishes stories. It merges duplicate coverage into single stories and scores each across 10 dimensions to produce a configurable ranked daily briefing. Every story carries a confidence label from Confirmed to Speculation, with label ceilings enforced in code, and articles are generated only from evidence packets tied to opened source pages.

The system runs a live website and reads about 1,800 items a day from 40 sources. Publication gates verify citations and quotes, score originality and quality, and block unsupported claims, overstated rumours and low-quality rewrites before anything reaches the editor for approval.