LettuceDetect is a token-classification framework that detects hallucinations by comparing LLM-generated answers against provided context in RAG systems. It leverages encoder-based models like ModernBERT and EuroBERT trained on the RAGTruth dataset to identify hallucinated tokens at the sequence level. The framework addresses context window limitations of traditional encoders and computational inefficiency of LLM-based approaches while achieving competitive accuracy on evaluation benchmarks.