David Len
Software developer. Building what future agents want.
7 posts · about 13 minutes of reading
Software engineer, quant trader.
Builds the QED tools: Novel Visual Studio, a writing IDE for dialogue-driven fiction, and Qron, a canvas for Claude-driven research.
Active on X, writes more non-commercial stuff on his personal site.
Shez and Kurumi are his AI characters for alternative expression. Perhaps one day Shez and Kurumi may get their own Vtube models.
- 2 min read
The errors that teach
An agent that loops is usually not a dumb model. It is a harness that warns up front and then abandons the model at the exact moment it needs help.
- 2 min read
Memory is not one thing
Our analysis builds a wiki of your story automatically. That is useful and it is also a ceiling, because a wiki knows everything at once and a reader does not.
- 1 min read
What we learned reading the classics
We pointed NVS at public-domain classics to build seed datasets: Romance of the Three Kingdoms, Journey to the West, Alice in Wonderland. The engine held up. The surprise was which books made good seeds, and the one number that predicts it.
- 2 min read
Particles rasterizing particles
A scroll effect that dissolves the page into sand, layered onto a hero that is already an animated particle field. The measurements were worse than the guess, and the fix was to want less.
- 2 min read
Your reader can see every tree and still miss the forest
We pointed the engine at a real 300-chapter story and it read every subplot faithfully, while the grand themes quietly slid into the lore shelf instead of the plot ledger. Here is why, and what it taught us about what "understanding a story" even means.
- 2 min read
Who knows what, and when
Dramatic irony is not a mood, it is bookkeeping: the gap between what a character knows and what the reader knows, tracked over time. So we tracked it, and the chart turned out to be a map of every suspense window in the story.
- 2 min read
The gap between what you meant to write and what you wrote
A planning board is a record of intent. A manuscript is a record of what actually happened. Put them side by side and the difference is not a mess. It is the single most useful thing you can see as an author: your frontier.