Weekly journal 2025-09-14~2025-09-22

For 15 days

A phenomenon is occurring where the cycle update interval is increasing even though there seems to be no significant external events.

Aug. 31 Team Mirai Hackathon

Only joined remotely for an hour at the beginning of the meeting because of a conflict with another matter.

What kept you busy for the next week?

Review of internal Idobata and discussion of next contract

Meeting on 9/4

Organizing a confusing story

9/5

Plurality's translated version released online today

One CC-BY book data and a number of articles, such as the Cybos formula, make good testbed data, and I'd like to take this opportunity for an interesting technical demonstration.

I want to create a "good summary" and to do so, I want to experiment with knowledge graph extraction (GraphRAG) systems.

It's also connected to last week's story about visualizing 20,000 data points in Kozaneba.

I think it's less than 20,000 cases if you simply visualize it in a scatter plot.

If I want to deal with the structure of chapters and the relationships between concepts without abandoning them, I want to use various features of Kozaneza rather than just scatter plots.

→This week I did a survey of GraphRAG related technologies, I'll try to include data from the Plurality book next week.

9/8 Monday Information extraction largely successful, try to add miscellaneous views

9/9 Release

Just one chapter for now.

Function to label edges, prototyped but subtle, so I'm thinking about designing it.

GPT's recommendation is a pretty intense rewrite route, skin deep that it's not a size that can be left to the AI to complete.

I was able to add links across chapters.

I got as far as deploying to Azure.

The way client builds is wrong and I have to do a static build every time.

---