Open research and tooling.

github.com/rororowyourboat/latent-topologies

Studying LLM latent-space topology using persistent homology and Hodge decomposition. Finds that the Hodge gradient potential recovers the animacy hierarchy — a linguistic universal — without any supervision.

github.com/rororowyourboat/mechanistic-interpretability

Head-level analysis of GPT-2-small: tracing how syntax, semantics, factual recall, and pragmatics are actually represented inside the network.

github.com/rororowyourboat/information_retreival

Information retrieval as epistemic architecture — notes on ontological commitments, bounded attention, and how to assemble context so downstream reasoning stays reliable.

github.com/rororowyourboat/structured_antagonism

A reasoning methodology for domains where ambiguity is expensive. Encodes the structural properties of rigorous thinking — design reviews, research plans, architecture decisions — directly into the process.