Example: https://blog.neutalk.io

NeuBlog was created through a fun experiment with me, and other members of the OpenAI community to find interesting ways to let an LLM search for information before posting.

For my contribution I created a simplified interface that exposed two functions. One to gather relevant news articles on Google News, and another that ingests the article by stripping LLM-irrelevant tags, and returning the content in markdown.

https://github.com/RonaldRuckus/article_nom

As can be noticed, this was created on December 2023. Over 1 year ago.

Upon creation, a team of community members noted the potential of this type of tool, and told me to continue the project to fruition.

The process of this service is simple and can be broken down into clear pipeline steps:

It's important to note that each LLM is given specific instructions on the company, the CTA, the target audience, and other general background information.

Gather Headlines Using a theme, a Google News search is conducted. The results of this search is sent to an LLM which determines which headlines seem to being in new, interesting, and potentially contrasting information

Gather Articles Once the headlines are gathered, each article is ingested, filtered, and converted into markdown

Distill Facts Each article is distilled into their facts by a capable LLM

Write Article Using the facts, and optionally keywords to target, a specific LLM is used based on the target audience (Creative VS factual. Educated VS Mass-appeal). The article is written in MDX format. It also creates prompts for imagery for a header and subsections.

Generate Images The prompts provided by the model for imagery are converted to images using both Veo2 by Google, and Dall-E by OpenAI. The reason behind this is that Veo2 currently has many limitations that Dall-E doesn't. Such as blocking any sort of prompt that implies the presence of people.

Link & Clean The image prompts generated are replaced with the file reference, and citations are placed to back up all facts with their respective article. Some artifacts created by the LLM are also cleaned up.

Move & Publish The files are moved to a specific location that is then immediately uploaded to GitHub. Triggering a website update.

https://blog.neutalk.io/blog/2025-the-year-ai-llms-transform-business-automation/

Ultimately, we have a fully capable automated blogging system that is up-to-date with recent news. Gathering numerous data-points from numerous sources, distilling it, and synthesizing it all into one potent article.

Warning: There still are some edge-cases with citations. For example, adding two citations together for whatever reason causes the renderer to spit out a different number:

> 2025 is set to witness massive adoption across various platforms and industries[^2^][^7^]

This for whatever reason turns into

> 2025 is set to witness massive adoption across various platforms and industries24