Medical Entity Extraction and Coding

Clinical phenotype annotation — build a gold-standard reference set (HPO) We're building a benchmark to evaluate an AI extraction pipeline for rare disease, and we need a human-annotated reference set to measure it against. Roughly 20 de-identified clinical documents — specialist letters, discharge summaries, genetics and imaging reports. You'll annotate clinical findings and map them to HPO, working blind to any model output. Full annotation guidelines and the submission template are attached. Please read the guidelines before applying — they cover span conventions, granularity, assertion status and the ambiguity log, and they'll tell you quickly whether this is your kind of work. The hard part isn't terminology, it's consistency. Negated and family-history findings are in scope. Genuinely ambiguous cases get flagged rather than silently resolved. Starts as a paid trial on 5 documents. We may run two annotators in parallel to measure inter-annotator agreement, so honest disagreement is a useful result, not a failure. Good fit: genetic counsellors, clinical genetics trainees, ontology curators, clinical informaticians, or anyone who has built an annotated corpus. Please don't apply if you'd use an NLP tool or LLM to generate annotations — that makes the benchmark circular. In your proposal, please answer: a specialist letter says "no evidence of seizures." How would you record that under the attached guidelines, and why does the answer matter for a benchmark?