
Argo
How do you trust a tool
that can be wrong?
A source-code auditing system that coordinates language models through focused analysis, independent validation and reviewable reports.
I designed the workflow, prompts and orchestration. Structured outputs, deterministic checks and reproducible artefacts make the results easier to question, inspect and improve.
Inside the project
Python, FastAPI, Pydantic and SQLite. Multiple model backends sit behind one runner interface; each stage has a distinct responsibility.
The software around the model can be tested with mocks. Measuring the model’s detection quality is a different problem. I keep those claims separate.
The project is independent of my earlier collaborative research at SDU.
