Scientific research, applied R&D, and technical notes from Chile
AI governance becomes useful when it is embedded into workflow design, approvals, logs, and evidence, not left as a static policy document.
A practical test plan for prompt injection, data leakage, RAG poisoning, tool abuse, excessive agency, and unsafe output handling.
RAG becomes enterprise-grade when retrieval is tied to source quality, user permissions, evaluation, and audit trails.
A scorecard for deciding whether an AI workflow should scale, stay in pilot, be redesigned, or be rejected.
Why vector storage cost, recall validation, and compression controls should be evaluated before a RAG program scales across the enterprise.
Before building a copilot or agent, map the process friction, decision latency, rework, and evidence needed to prove value.
A research-backed note on using scale-invariance tests as part of AI system evaluation, falsification, and deployment discipline.
A research note on recursive structure, compression, and why complex AI systems need falsifiable evaluation rather than abstract certainty.
EigenKV explores KV-cache reduction for long-context AI workflows where memory cost, latency, and quality must be evaluated together.
EigenWeights explores model footprint reduction for controlled deployments where latency, infrastructure limits, and quality thresholds matter.