Research local Large Language Models that can fit within 16GB VRAM, respond within 5 seconds (the less the better), and are available for commercial use. You will need to research both a back end to run the LLM and the LLM to use.
The LLM will need to be able to respond in a JSON or YAML format, accurately say items from a list and be able to give yes/no answers.