Publications
Peer-reviewed work, conference abstracts and editorials on confidential AI for regulated fields — each one readable in full, right here.
5 publications
Gabriel Tremblay (Untraceable AI) · Sharada Harricharan (Frontier HEOR)
A controlled comparison of de-identification methods — censorship, substitution, and Untraceable's sequential and semantic cloaking — measuring how much AI writing quality each one preserves.
Why confidentiality — not capability — has been the real barrier to AI adoption in HEOR, and how cloaking lowers it.
Proposes a writing-quality benchmark for large language models in HEOR — the measurement layer the field has lacked.
A controlled, multi-model test of whether retrieval-augmented generation over a Canadian HTA knowledge layer improves AI writing quality in HEOR.
A Delphi-style framework for benchmarking LLM writing quality in HEOR, demonstrated as a Canadian proof-of-concept.