Real-world LLMOps: Two case studies in healthcare AI deployment

Here's what nobody tells you about getting large language models into production: it starts with user stories. I know, I know. User stories feel like something the product team is always pushing on you. But hear me out.

When we started working with Boost Medical Group, an AI-powered personal injury company, we didn't begin with the latest GPT model or debate fine-tuning strategies. We started by sitting down with the CEO and the chief product officer. We interviewed customers. We mapped workflows. We got our hands dirty understanding the actual, real-world processes these tools would need to support.

The same approach was applied when we built an AI neuropsychiatrist for Sema Therapeutics. Our client, Dr. Katie Walker from Revitalist, had a crystal-clear request: she wanted a robot version of herself.

Not in some sci-fi way, but as a practical solution to being overwhelmed with patient assessments. She needed something that could handle depression and anxiety screenings without burning out her human staff.