vishakhasadhwani/llm-deployment-demo ? reverse-engineered prompt

Build me a private LLM chat app that I can run on my own GPU, either locally or on an AWS EC2 GPU instance. I want a simple web chat interface where I can talk to a model like Qwen2 1.5B Instruct, with the model running on the machine itself so no data goes to a third party and I do not need API keys.

Please package everything so it can run in one Docker container, with the backend model server starting first and the chat UI coming up after it is ready. The app should show a clear offline or loading state while the model is still starting, then let me refresh and start chatting once it is live.

It would be great if I can change the model easily with an environment variable, and if the setup works with common instruct or chat style HuggingFace models. If you need current setup details, look them up online and make the app work cleanly.