QubitLogic

The Challenge

Quantum computing coverage has a credibility problem: vendor announcements are reprinted as achievements, and a reader cannot tell a demonstrated result from a roadmap promise. Building a site that does better means solving two hard problems at once. The editorial one is that AI writes plausible, confident, well-formatted claims that nobody made - the exact failure the site exists to call out. The operational one is that one person cannot manually monitor arXiv, vendor blogs and lab feeds daily, so without automation the site goes stale within a month.

Our Solution

A four-project .NET 8 solution built around one rule: the model extracts and evidences, code decides, a person approves. Ingest polls arXiv and vendor feeds, deduplicates across outlets in three stages, and labels relevance rather than discarding - a filter drops records that were never written down, so the mistake stays invisible. An AI drafter extracts four analysis fields and reports which rubric signals it can evidence with a direct quote; it never chooses a hype rating. Scoring is arithmetic in code, so identical evidence always yields the same band and swapping AI providers cannot change a published rating. Both providers share one prompt and one parser for the same reason. Auto-publish still enforces an editorial floor: no primary source link, or a 'what they did not achieve' field under 20 characters, holds the story as a draft. Eight lessons each carry a purpose-built canvas interactive, with the physics computed from constants rather than fitted to look right. A roadmap watcher re-fetches vendor sources on a schedule and proposes changes with a quote for review - it is forbidden from editing a row, because every row is a dated commitment attributed to a named company.

The Results

Live at qubitlogic.io on self-hosted IIS and SQL Server. Eight lessons with eight interactives, a monitored news pipeline, a sourced roadmap tracker, generated sitemap and robots, structured data, and automated Facebook publishing. The engineering result worth naming is the test harness: 215 static checks plus a stub-canvas runner that executes every visualisation, because the renderer catches its own errors and removes the node - a broken interactive looks like empty space, not an error. Several checks were rewritten or deleted after firing on correct code, on the principle that a check which reports a false failure is worse than none. No traffic or revenue claims are made here: the site launched in August 2026 and any figure quoted this early would be the kind of number the site itself exists to question.

Project Details

Tech Stack

Tags