Personal AI News Desk

I built this personal AI News Desk to solve a very specific problem: I had more useful newsletters, blogs and research feeds than I had time to review.

Rather than create another feed to scroll, I wanted one calm, structured briefing that could help me identify what was genuinely worth reading while keeping source selection and final judgement in human hands.

At a glance

• Use case: Information triage and content monitoring
• Inputs: Selected RSS feeds and email newsletters
• Outputs: Daily brief, top-ten articles, research-paper pick and topic trends
• Built with: Codex and a local workflow
• My role: Workflow design, prototyping, testing, preference design and iteration
• Status: Working personal prototype

The problem

Following AI research and industry developments had become a fragmented manual process. Useful material was spread across email newsletters, publications, independent blogs and RSS feeds.

Bringing everything into one place would not have been enough. A larger pile of links would still require manual scanning and create another source of noise.

The real problem was how to collect, structure and prioritise the information without handing all editorial judgement to an opaque recommendation feed.

What I built

• Selected material gathered from RSS feeds and email newsletters
• Incoming items processed into a consistent structure
• One concise daily briefing
• Ten ranked articles
• One highlighted AI research paper
• An emerging-topic and trends view
• Bookmarks for later reading
• Preference feedback to improve future selections
• Source controls so I can add or remove inputs
• Direct access to original material for review
• Final control retained over what is useful

Workflow

Selected sources

RSS feeds and email newsletters.

Collection and processing

Incoming items are gathered and converted into a consistent structure.

Preference-based ranking

Content is assessed against the subjects, sources and characteristics I value.

Daily output

Briefing, top-ten articles, research paper and topic trends.

Human feedback

Review, bookmarking, source checking and preference refinement.

Why Codex

Codex helped me move from "I wish I had a tool for this" to a working local prototype.

I used it to explore the workflow, structure the application, troubleshoot issues and iterate on features. My contribution was not simply providing one prompt and accepting the result. The work involved defining the problem, making decisions about inputs and outputs, testing what was useful and refining the system around my own reading behaviour.

This is an example of how I use Codex as a practical building partner: the tool supports development, while the human remains responsible for purpose, judgement, boundaries and quality.

Human review and responsible use

The News Desk is designed to support attention, not replace it.

AI-generated rankings and summaries may omit context or overstate the importance of a story. For that reason, the workflow retains direct links to original sources and keeps human review central to the process.

I choose the inputs, review recommendations, open the source material and decide what deserves further investigation.

What the build demonstrates

• Moving from an everyday frustration to a working AI-assisted tool
• Turning messy multi-source inputs into a structured workflow
• Designing for human review rather than full automation
• Using preference feedback to improve relevance over time
• Building a tool around an actual user, in this case me
• Documenting lessons so the underlying pattern can be applied elsewhere

What I learned

The most important design decision was not which model or feature to use. It was deciding what a useful daily output should look like.

A successful information tool should not simply process more content. It should reduce cognitive load and help the user make a clearer next decision.

I also learned that preference-based systems need visible feedback loops. If the recommendations are not useful, the user needs a simple way to correct them.

Current status and next steps

The News Desk is a working personal prototype and continues to evolve.

Next steps include testing recommended-source discovery, clearer explanations of why individual items were selected, and improved topic tracking over time. New sources will continue to require human approval before they become part of the workflow.