NewsForge
AI newsroom production pipeline with multiple specialised agents and a human-in-the-loop editorial review path.
Definition
What is the NewsForge case study?
NewsForge is a LoneSock-built AI newsroom production pipeline. It uses a set of specialised agents covering source monitoring, story detection, drafting, fact verification, image selection, headline optimisation, SEO tagging, translation, and distribution, with a human-in-the-loop editorial review path. It is operated as a KaritKarma platform product.
The Challenge
What we were solving for.
A news operator wanted to expand editorial coverage across language editions without scaling the editorial team proportionally. Manual article production could not keep pace with the cycle across politics, sports, business, technology, and entertainment.
AI-generated copy had to clear the editorial bar that human-written articles clear. Fact verification, source attribution, and tone consistency could not be relaxed for speed.
The pipeline had to integrate with existing CMS infrastructure without forcing a migration. Editorial control and the final publish decision had to stay with humans.
What We Built
Technical architecture.
A multi-agent editorial pipeline in which each agent owns a defined editorial function: source monitoring, story detection, drafting, fact verification, image selection, headline optimisation, SEO tagging, translation, and cross-platform distribution.
A scanning surface that watches a large source set for breaking stories and trending topics, with detected stories flowing through the agent set into draft form within minutes rather than hours.
A human-in-the-loop approval surface that lets editors set per-section policy (auto-publish for low-stakes content, mandatory editorial review for sensitive topics) and shows agent confidence scores, source chains, and suggested edits.
CMS adapters covering WordPress, Drupal, Ghost, Strapi, and Contentful out of the box, plus a REST API for custom platforms.
Key Numbers
Attributed, honest figures.
We publish numbers that the engagement supports. Unverifiable marketing metrics are not on this page.
Technology Stack
What we built it with.
Outcome
What the engagement delivered.
Editorial throughput across editions scales without proportional editorial headcount growth.
Multi-lingual publication runs from a single source story, with translation that preserves tone and cultural context rather than producing a literal translation.
Editors retain the final publish decision and the system surfaces what they need to evaluate it: confidence, sources, suggested edits.
Questions
FAQ for the NewsForge engagement.
What did LoneSock build for NewsForge?
LoneSock built the NewsForge multi-agent editorial pipeline, the editorial review surface, the multi-lingual translation stage, and the CMS adapters for WordPress, Drupal, Ghost, Strapi, and Contentful. The product is operated as a KaritKarma platform offering.
What tech stack was used?
Python for the agent orchestration, .NET 10 for surrounding services, Groq for the LLM tier, Next.js 16 for the editorial surface, and PostgreSQL for state. Docker for deployment.
Does NewsForge auto-publish?
Only where the operator authorises it. Per-section policy can be set to auto-publish for low-stakes content or to require editorial review. Sensitive topics ship into a review queue regardless.
What CMS platforms are supported?
WordPress, Drupal, Ghost, Strapi, and Contentful through built-in adapters. Custom platforms are supported through a REST API.
Is the platform still maintained?
Yes. NewsForge is an active KaritKarma platform product with ongoing development from LoneSock.
Working on something similar
We have shipped this class of system before. Tell us what you are dealing with.
$20/hr time and materials, fixed-price milestones, or retainer. Source code is yours.