From ML Tooling to Analytical Governance: Recent Updates to KMDS

A developer has been refining KMDS, a framework for building repeatable and auditable machine learning systems, with recent updates focusing on analytical governance. New features include a Feature Advisor service that recommends feature engineering strategies and a Design Governance framework that captures decision points and generates implementation guidance. The long-term goal is to preserve analytical knowledge as reusable assets through knowledge graphs, ensuring institutional knowledge is retained even when team members leave.

Over the last few months I've been refining KMDS, a framework for building repeatable and auditable machine learning systems. The original motivation behind KMDS was simple: Many machine learning projects fail long before model selection becomes important. Teams struggle with questions such as: Most organizations answer these questions at some point. The problem is that the answers often disappear into notebooks, documents, tickets, or the memories of individual contributors. KMDS is an attempt to make these decisions explicit, structured, and reusable. Recent updates have focused on moving beyond workflow automation and toward analytical governance. The workflow begins with semantic tagging and metadata generation. Rather than immediately building features or training models, the system first attempts to understand: The goal is to establish a semantic foundation before modeling begins. One of the new additions is a Feature Advisor service. Given metadata and project context, the advisor recommends feature engineering strategies for non-numeric attributes. Examples include: The objective is not automatic feature engineering. The objective is to provide design guidance and rationale that helps practitioners make better decisions. A second addition is a Design Governance framework. Machine learning projects contain many decision points: The Design Governance layer acts as a design-time advisor that captures these considerations and generates implementation guidance. The output is a structured design blueprint that can be reviewed by humans or supplied to AI coding assistants. Perhaps the most important change is an increased emphasis on preserving analytical knowledge. The long-term goal is not simply to create models. It is to create reusable analytical assets. Using KMDS tooling, project artifacts can be transformed into a knowledge graph representing: This creates a queryable representation of the analytical lifecycle. Most organizations already have documentation. What they often lack is accessible institutional knowledge. Critical analytical decisions are frequently distributed across: When people leave, much of that context leaves with them. My view is that the real asset is not the agent. The real asset is the structured analytical knowledge that the agent can access. If the knowledge is preserved independently of any specific model, tool, or LLM, organizations retain ownership of their analytical reasoning and can recreate capabilities as technology evolves. The broader goal of KMDS is to make machine learning systems: Recent work has focused on feature governance, design governance, metadata-driven workflows, and knowledge graph generation. Future work will continue exploring how analytical context can be captured and preserved as a first-class artifact rather than an afterthought. I would be interested in hearing how others are approaching analytical governance, reproducibility, and knowledge preservation in their own machine learning workflows.