Path consistency for latent world models
A self-supervised mechanism that penalizes disagreement between horizon-conditioned direct prediction and autoregressive rollout, aimed at reducing compound error in latent rollouts. Designed as an importable component rather than a standalone architecture.
Acoustic early warning, and where it breaks
A Random Forest classifier on MFCC features with local inference, reaching strong held-out metrics that then failed on real recordings while misflagging ambient noise. The generalization gap, not the accuracy number, is the finding.
An AI teaching assistant that plans its own courses
Describe what you want to learn, add your own notes and files, and Doc plans a course as a map of topics, then writes lessons and tasks, grades answers, and schedules reviews. FastAPI, PostgreSQL with pgvector, React, models through OpenRouter.