Recent advancements in large-scale language models have paved the way for intelligent systems capable of understanding and generating domain-specific text. BioGPT, a generative transformer model pre-trained on biomedical literature, offers substantial improvements in tasks like question answering and knowledge extraction in the healthcare domain. In this paper, we present MedBot, a biomedical based on a decoder-only transformer (BioGPT) using the BioASQ dataset. This system achieves a BLEU-2 score of 0.42 and reduces perplexity to 8.3 (compared to 12.1 for BioBERT baselines), demonstrating significant improvements in response quality. The model generates answers with 73% falling within the optimal 35–55 token range for medical responses, and medical professionals rated 82% of outputs as clinically adequate. The system is designed to process medical questions and provide context-aware, accurate responses, making it useful in medical education, clinical research, and patient engagement platforms. Here the model will be described through architecture, training methodology, and design choices, and compare its output against baseline models to demonstrate its efficacy. the results indicate that prompt engineering and domain-specific fine-tuning improve answer relevance by 18%, making MedBot a promising step toward intelligent biomedical assistants.