dchisholm125

https://github.com/dchisholm125/graph-oriented-generation

I spent months running experiments on what language models do when they encounter inputs outside their training distribution — random phonemes, invented morphemes, Wierzbicka's semantic primitives.The finding that surprised me: Language model behavior follows a reproducible taxonomy (Synthesis, Collapse, Overflow, Metacognition, Linguistic Drift). It's not random noise — it's classifiable.The...

https://github.com/dchisholm125/graph-oriented-generation

LLMs are better at being the "mouth" than the "brain" and I can prove it mathematically. I built a deterministic graph engine that offloads reasoning from the LLM. It reduces token usage by 89% and makes a tiny 0.8B model trace enterprise execution paths flawlessly. Here is the white paper and the reproducible benchmark.

https://github.com/dchisholm125/Talk2Code

Talk2Code was my endeavor into building something useful, but I thought it would take MUCH more than 150 lines of code... this is just stupid simple, and it SHOULD be shared. Check out my video walkthrough on Reddit too: https://www.reddit.com/r/SideProject/comments/1r5wvzv/i_buil...