01
the system gathers fresh sources into a queue
Product case study
AI News Creator 2.0: content pipeline with human review
AI News Creator 2.0 shows how to build a content workflow with AI without chaos and without blind autopublishing: from sources, through screening and research, to a prepared draft awaiting approval.
Key implementation elements
ingest of fresh sources into one flow
automatic screening with human topic selection
separate collect_sources and verify_facts stages
research packs as context bundles for writing
prepared state instead of automatic publication without approval
Problem before implementation
Problem before implementation
In content pipelines the hardest part is not writing itself, but organizing sources, topic selection, fact verification, and approval before publication. Without that, AI quickly creates noise instead of a predictable process.
sources and topic ideas appear in different places
no explicit stages from screening to ready draft
risk of mixing research, drafting, and publishing into one step
need for manual approval instead of uncontrolled autopublishing
What we built
What we built
We built a Gemma-first editorial pipeline with ingest, screening, human selection, source collection, fact verification, research packs, and final draft preparation locally or through an external provider.
ingest of fresh sources into one flow
automatic screening with human topic selection
separate collect_sources and verify_facts stages
research packs as context bundles for writing
prepared state instead of automatic publication without approval
Process after implementation
How the process works after implementation
From input data to a cleaner outcome. Below is a shortened view of the process after implementation.
02
the model performs the first screening of topics
03
a human decides which topics move forward
04
for selected topics, the pipeline gathers sources and verifies facts
05
the result is a research pack and a draft prepared for editorial decision
Before / after
How the process changed
The table shows the main differences between manual work and the process after implementation.
| Before implementation | After implementation |
|---|---|
| research, selection, and draft merged into one chaotic stage | separate stages from ingest to prepared draft |
| AI runs without a clear human control point | human selection and approval are built into the process |
| sources and facts are hard to reconstruct later | the research pack preserves context for further work |
| publication easily mixes with experimentation | the draft goes to manual decision instead of autopilot |
Business outcome
a more predictable process for working with content and sources
clear separation between screening, research, and final drafting
lower risk of chaotic autopublishing
easier use of AI as process support instead of a magic shortcut
Technologies
lokalne modele GemmaLM Studio / LMS CLIpipeline CLIresearch packi i weryfikacja faktówworkflow z człowiekiem w pętli
What can be implemented in a similar way
What can be implemented in a similar way
These are examples of processes that can be organized with a similar approach: start from one concrete problem and a clear data flow.
research assistants for marketing or editorial teams
pipelines for selecting and organizing sources
internal tools for building drafts with human control
AI supporting classification, research, and operational content drafting
Related services
This type of implementation can be connected with MorenaTech's core areas
If a similar process still runs manually or is scattered across files, it can be connected with automation services, Google Workspace, or further process development.
Final CTA
Want to organize a content or research workflow in a similar way?
If your team is testing AI but still lacks structure between sources, verification, and approval, it is worth first building a simple workflow with a human in the loop.