AI Agent That Turns Scattered Health Records into a Trustworthy Patient Summary

Design an AI agent for healthcare that gathers a patient's medical history and produces a summary of it. Cover both the product and the system behind it. ```hint Start with the reader A summary is only as good as the decision it supports. Decide who reads it and when, then ask what a wrong or missing line would cost that person; the answer shapes the content, the format and the safeguards. ``` ### Clarifying Questions - Who reads the summary: a physician preparing for a visit, an emergency clinician seeing the patient for the first time, a care coordinator, or the patient? - What sources may the agent use: structured data such as diagnoses, medications and lab results, free-text clinical notes, scanned documents from other providers, or all of these? - Is the agent read-only, or may it take actions such as drafting orders or sending messages? - Which privacy rules apply (for example HIPAA in the United States), and where may patient data be processed? ### What a Strong Answer Covers - A chosen reader and moment of use, and the summary content and format that serve that decision. - The path from fragmented records to a summary: data access, normalization, retrieval, generation and citations back to the source. - Safety for a high-stakes setting: grounding every statement, surfacing conflicting or missing data, keeping a clinician in the loop, and limiting what the agent can do. - Privacy, security and compliance: consent, access control, audit trails and where the model runs. - How success is measured, covering accuracy and omissions as well as time saved and adoption. ### Follow-up Questions - A clinician reports a summary that left out a severe drug allergy. What do you do right away, and what changes in the system? - How would the product change if patients read summaries of their own history? - How would you show that the summaries improve care rather than only saving time?