BALM (Binding Affinity Language Models) is a deep-learning framework that predicts protein-ligand binding affinity (pKd) by fine-tuning pretrained protein (ESM-2) and ligand (ChemBERTa) language models with PEFT adapters and projecting their pooled embeddings into a shared space, where binding affinity is read out from the cosine similarity of the protein and drug projections.