Kowalski Analysis

We were inspired by our struggles with maintaining long term AI-generated coding projects, as well as our fascination with graph algorithms and heuristics!

We combined these two areas in our project, Kowalski Analysis! Inspired by the one and only super-intelligent Penguin from Madagascar.

We use two never-before used static code analysis metrics: Impact - the likelihood for a piece of code to cause adverse affects in the codebase and Susceptibility - the likelihood for a piece of code to be impacted by a stray edit.

We calculate these metrics based off of the Afferent and Efferent Coupling of different sections of code, which basically just means who and how often they call each other.

The result, is an MCP integrated tool you can add to any AI Agent to provide it with important metrics to make design decisions that it never could have had before.

Furthermore, we created an awesome, Obsidian inspired graph visualization that allows you to see how your codebase grows like a tree, with harder trunk wood that should not be edited often, and leafy greens that are easy and risk free to edit!

We primarily built our tool in python with FastMCP for integration, but also deployed a version to the Fetch.ai agentverse using their native LLM for even more insights on our data!

Our largest challenge was determining the new metrics, as they definitionally create code that minimizes how much additional code you have to change when making an edit to any file.