Work / Healthcare / Private Analytics
Letting staff ask questions about private data safely
A question-and-answer tool that uses AI to understand the question without giving the AI the private records.
Useful answers were locked behind technical database knowledge
Giving an outside AI service unrestricted access to private records was not acceptable.
AI translates the question without receiving the underlying records
The query runs inside the practice’s controlled environment with narrow permissions.
Staff can explore business data while the practice retains custody
Conversational access becomes practical without treating privacy as an afterthought.
The answers existed, but only a technical person could retrieve them
The practice had years of useful information, but answering a new question required someone who understood the database. A chat-style tool could make that information easier to use, but simply sending private records to an outside AI service created an obvious privacy problem.
The AI needed the map, not the contents
The AI needed to know what kinds of information were available and how they were organized. It did not need to see the actual customer records or run searches against them itself.
I separated understanding the question from looking up the answer
When a staff member asks a question, the AI sees the question and a description of how the database is organized. It writes the search instructions, but the search itself runs inside the practice’s controlled environment. Only the final result is shown to the user. Customer identities are replaced with anonymous identifiers during processing.
The AI could help without holding the data
The key was giving the AI one narrow job: translate an ordinary-language question into database instructions. Access to the records, the actual search, and the link back to real identities all remain under the practice’s control.
Staff gained a simpler way to explore the information
Staff can ask questions conversationally instead of learning a database language, while the practice keeps control of the records and the connection to real people.
Privacy became part of the design, not a promise added later
The prototype showed that an AI feature can be useful without being allowed to see or control everything around it.
The AI does one useful job and no more
The AI writes the search instructions. The practice’s own system decides what those instructions can reach and keeps custody of the private information.
Keep exploring
See what else I have built.
Browse more systems, tools, and consulting engagements across different industries and kinds of problems.