AI Workstation & Cloud Development

I need an end-to-end setup that lets me train, fine-tune, and run a custom AI focused on automated customer service. All data involved is text— chat logs, emails, knowledge-base articles, and similar material— so the environment must be optimised for natural-language processing from ingestion through deployment. While text is the priority, I would like the architecture left flexible enough to add numerical or image pipelines later without a complete rebuild.

Scope of work
• Specify and assemble a local workstation (GPU, RAM, storage) that can handle large-scale language-model training and experimentation.
• Stand up a complementary cloud stack—-preferably on AWS or Azure—-for elastic training jobs, model versioning, and low-latency inference behind an API.
• Build the data pipeline: secure transfer from my on-prem sources, cleaning, tokenisation, labelling, and continuous updates.
• Implement model fine-tuning (e.g., PyTorch or TensorFlow, Hugging Face transformers) with clear scripts so I can retrain as business content evolves.
• Deliver a chat/FAQ service that plugs into my website and internal tools, complete with fallback rules, confidence scoring, and logging so I can review answers.
• Document every step— hardware diagrams, infra as code, setup scripts, and how to trigger training or roll back a model.

Acceptance criteria
1. Workstation image and cloud instance both spin up and run the same codebase.
2. A demo chatbot answers test questions pulled from my text corpus with at least 90 % accuracy on an agreed validation set.
3. Full documentation and commented code are provided in a private repo.
4. I can reproduce training and deploy a new model end-to-end in one command.

Please respond with examples of similar AI or NLP infrastructures you have designed or deployed— screenshots, repos, or live demos are all welcome.