A research workspace built around inspection
DAAF wraps an AI coding agent in a Docker-based quantitative research environment. It keeps analysis in Python or R files, records decisions and outputs, and supplies workflows for data profiling, planning, analysis, review, reporting, and reproducibility checks.
The useful difference is not autonomous research. DAAF is designed for a researcher who can choose the method, inspect the code, and challenge the result. A practical first project is to load one documented public dataset and ask for a profile, then compare every generated variable description with the source documentation before doing any modelling.
It is deliberately heavier than a single skill
Installation requires Docker plus access to a supported model provider. The project warns that a full API-backed analysis can be expensive, and its own documentation says expert review remains necessary because models can still hallucinate or cut corners.
DAAF was too specialised and framework-heavy for this week’s email, but those are not catalogue rejection reasons. It has a concrete use, public source, current release, and documented limitations, so it belongs here for readers who do quantitative research.