Embeddings that link medical terms across English and 12 Indian languages.
A medical embedding model fine-tuned from bge-m3 on more than 18 million term pairs aligned to SNOMED CT and UMLS. Works in native and romanized scripts, and copes with abbreviations, misspellings and colloquial terms - 'sugar problem' lands near diabetes mellitus.
Benchmarks
| Metric | Value | Dataset |
|---|---|---|
| Recall@1 (higher is better) | 0.7206 | Eka-IndicMTEB |
| Recall@5 (higher is better) | 0.8512 | Eka-IndicMTEB |
Install
Accept the terms on Hugging Face, then download
hf download ekacare/parrotlet-e Links
Topics
embeddingssemantic-searchmedical-codingentity-linkingmultilingualindic