Deb Bose
Research • Notebooks • Papers • Projects
I help APRA-regulated risk and data teams build continuous AI assurance tooling — the pipelines, controls, and observability that make ML and LLM systems auditable under CPG 235 and BCBS 239, not just demoable. I also take on quantitative research and end-to-end AI/ML delivery: from framing the question through to a system running in production.
Eighteen years shipping data platforms and quantitative systems in financial services, most recently leading data modernisation at Macquarie Group. GARP-certified in Risk and AI (RAI) and CFA Level 1, so neither the assurance nor the finance is bolted on. If you have models in production and a regulator who wants evidence they behave — or a research question that deserves a real system behind it — let’s talk.
APRA’s April 2026 letter now expects second-line functions to have tooling to independently assess AI systems. That tooling is what I build.
The rest of this site is a working archive — not a feed. It exists because I think in public.
Everything here is oriented toward truth-seeking through analysis: quantitative reasoning where possible, first-principles thinking where necessary, and skepticism toward narratives that don’t survive contact with data.
Work with me
I take on a small number of engagements where the problem is concrete and the stakes are real:
- AI assurance and model governance — continuous validation, drift and calibration monitoring, and audit trails that satisfy CPG 235 / BCBS 239 reviewers.
- End-to-end AI/ML delivery — scoping, data, modelling, and deployment, owned through to a system that runs and is maintainable.
- Quantitative research — portfolio analytics, economic time-series, and financial modelling, grounded in CFA Level 1 fundamentals and turned into reproducible systems rather than one-off notebooks.
- Data platform modernisation — DBT, Iceberg, Redshift, and observable pipelines, built so the lineage and controls are first-class, not bolted on.
If that maps to a problem you have, email bose.debasish@gmail.com or read more about my background.
What you’ll find here
Articles
Essays, market notes, and technical deep dives — written to clarify, not to perform.
Notebooks
Runnable research with code, charts, and assumptions made explicit.
Exploratory by design, not polished marketing artefacts.
Papers
Drafts, PDFs, and reference material — including work that is incomplete, evolving, or deliberately unresolved.
Projects
Pointers to my GitHub work: tools, models, and systems built to answer specific questions.