One narrow question.

ContextRx is an independent research project exploring the use of NLP and machine learning to detect contextual word errors in pharmaceutical documents and drug labels — real words, correctly spelled, wrong for the clinical context they appear in.

It has gone through multiple architectural iterations, from rule and n-gram baselines through statistical language models to learned contextual representations fitted on a curated corpus of FDA-approved label documents.

It currently operates as a production system scanning FDA-approved drug labels and identifying errors that pass through existing pharmaceutical quality systems, including regulatory review. That result — that this error class survives every layer of review and reaches the published label — is the finding the project is built around.

Why this class of error

Spellcheckers were designed for the office document, and general-purpose grammar tools for everyday prose. Neither is built to know that "patents" in a dosing instruction, or "temperate" in a storage condition, is a clinically wrong word. The error is invisible precisely because it is spelled correctly.

Scope, Independence & IP Architecture

ContextRx is an independent research project independent of any commercial sponsor, offering modular, offline-first NLP assets including custom context models, token-scoring algorithms, and a curated corpus of normalized FDA-approved labels engineered for system integration or technology transfer.

Peer review, replication attempts, methodology critique, and collaboration are all genuinely welcome.

The ContextRx case hub documents the two current human-verified examples and links to the full case analyses.