"GipsyRanger team on IBM Bob 2.0 hackathon Hackathon"

Team Idea

What I'm building: Codebase Autopsy - a CLI that audits a Python repo for "architectural drift." It walks the codebase with ast, builds a full import-dependency graph (flagging cycles and high-fan-in hub modules), then for every module with a docstring, asks an LLM to judge whether the code's actual behavior - its functions, classes, and what they call - still matches what the docstring claims it does. Each module gets a 0–10 drift score with concrete evidence, so instead of trusting stale comments, you get a ranked report of exactly where the documentation and the implementation have quietly diverged. Who I need: I'm solo, so what would help most is a teammate with experience wiring up LLM CLIs reliably (prompt design + enforcing strict JSON output), plus anyone willing to point it at a genuinely messy real-world repo to stress-test the drift detection and help tighten the report/UI.

Team Members

Fortune Ramafemo

GipsyRanger

Software Engineer

Submission

Codebase Autopsy

Codebase Autopsy scans a Python repo, maps its import graph, and asks an LLM to judge whether each module's actual behavior still matches what its own docstring claims, then ranks the worst offenders by drift score.