Solving Emergency Department Triage with Small Language Models: Why Large Commercial Models Fail and How Specialized Training Achieves Clinical Accuracy
Research paper demonstrating how specialized small language models outperform large commercial LLMs for emergency department triage classification, achieving clinical accuracy through domain-specific training.
Read article →
Transformer Architectures for HealthIT and Finance: An Architect's Field Guide
Encoder-only, encoder-decoder, and decoder-only Transformers — architectural reasoning for HealthIT and Finance practitioners. When to use BERT, T5, or GPT-class LLMs for clinical NER, ICD coding, summarization, agents, and trading AI. Written by Vadzim Belski, Head of AI at ScienceSoft.
Read article →NeMo Text Processing — Arabic Emirati
Fork of NVIDIA NeMo text normalization adding Emirati Arabic (ar_ae) dialect support. WFST-based normalization for numbers, dates, currencies, and Gulf Arabic-specific entities.
Read article →