Surpassing vLLM with a Generated Inference Stack - Tech Sentiments