Case Study — Nemera — AI Internal Chatbot
A secure internal chatbot that lets a medical-device manufacturer query decades of historical product data in natural language. Led prompt engineering and input sanitization; cut API costs 40%.
At a glance
Nemera, a medical-device manufacturer, had decades of historical product and test data locked in documents that were slow to search by hand. They wanted engineers to ask questions in plain language and get grounded answers — without exposing sensitive data or running up unbounded LLM costs.
On a Georgia Tech senior-design team for the client, I owned the LLM integration — prompt engineering, input sanitization, and the conversation and data-scoping architecture.
Built a secure internal chatbot over the company's historical medical-device data using an LLM API.
Designed a data-scoping architecture that grounds each response in the relevant document subset, so the model answers from the right context instead of hallucinating across everything.
Led prompt engineering and input sanitization — hardening against injection and trimming wasted tokens, which cut API costs 40%.
Designed the multi-turn conversation flow so follow-up questions stay grounded in the same scoped context.
Delivered a working secure chatbot for the client that turns slow manual document search into natural-language queries, with a 40% reduction in API cost and responses scoped to the relevant data.
With LLMs, most of the engineering is around the model, not in it — scoping the context, sanitizing the input, and shaping the prompt mattered far more than the model choice. Cost and safety are design constraints you architect for, not afterthoughts.