Research¶

Publications from the KI7MT Sovereign AI Lab, documenting the IONIS project's methodology, infrastructure, and lessons learned.

Every paper is formally published with a permanent DOI on Zenodo and collected under the Agentics Labs community. The Zenodo record is the canonical, citable version — please cite the DOI, not this page.

Published papers¶

The Dialectical Engine¶

Multi-Agent Falsification in Scientific Software Development

How five AI agents and a human domain expert used dialectical analysis to debug a neural network — and discovered that the failure record was more valuable than the model.

From Nanometers to Neurons¶

Applying Semiconductor Process Control to Machine Learning

CLCA / 8D methodology from semiconductor manufacturing applied to neural network development — systematic defect isolation across 16 model versions, each dead end with a documented root cause.

Sovereign Infrastructure for Citizen Science¶

Building a Self-Hosted Research Lab for Radio Propagation Analysis

How a self-hosted lab processes 14.68 billion rows on commodity hardware — 24.67 Mrps ingestion, 60 GB/s query throughput, 30 tok/s local LLM inference — at roughly 1.3% of equivalent cloud cost.

MCP Beyond AI¶

80 Tools, 14 Billion Rows, Zero Cloud — The Full Stack

The Model Context Protocol isn't an AI thing — it's JSON-RPC over stdio, the integration pattern Unix has used since 1973. What happens when you wire 80 MCP tools, 14 billion rows, a local 70B LLM, and zero cloud into one sovereign stack running from a house in Idaho.

Who Says Hermits Don't Have Friends¶

The Human Story Behind a Sovereign AI Lab

The non-technical story of building IONIS — and the framework that governs the agents themselves — told through the lab's "Collected Moments": the caught-and-owned mistakes, where the discipline is that nobody is right and the system doesn't need them to be.