Research & Writing

I write and publish at the intersection of AI engineering, evaluation methodology, and regulated industries.

Published work

Ongoing writing

I publish ongoing essays on AI engineering, evaluation methodology, and Islamic finance technology at qals.substack.com.

Current research

I'm currently building an evaluation platform for a specific class of LLM failure mode — silent source-blending, where models synthesize authoritative sources that have actually split on an issue and present the result as consensus. Think of it as an LLM answering a legal question by mixing reasoning from one federal circuit and holdings from another, presenting the synthesis as "the law" without flagging the split. The platform is being instantiated in Islamic jurisprudence as a proving ground — the jurisdictional structure there is unusually clean and well-documented — with the methodology designed to generalize to other authoritative-source domains: legal circuit splits, medical guidelines across regulatory bodies, and regulatory standards across regimes. Co-authored with Dr. Ziyaad Mahomed. A working paper is forthcoming on SSRN.

Speaking & teaching