HectorAI: assistente médico para prontuários com AI

Abstract

The medical record is a legal, confidential, and scientific document that supports continuity of care, but producing it consumes a relevant part of the physician's workday and contributes to professional overload. This report aimed to develop and document HectorAI, a web application that automates the production of SOAP clinical notes from the audio of consultations conducted in Brazilian Portuguese. The work comprised a literature review on clinical documentation and artificial intelligence assistants, requirements elicitation, architecture modeling, and incremental prototype development. The solution was implemented with Ruby on Rails, asynchronous processing, automatic speech transcription by Whisper, and structured text generation by GPT-4o. The result was a functional prototype that allows physicians to register patients, schedule and record consultations, monitor processing, review the generated note, and approve it for registration. The application also records changes made by the physician to guide subsequent generations. The proposed architecture therefore integrates audio capture, transcription, and clinical documentation into a single workflow and provides a technical basis for reducing manual documentation work. Evaluation in a real clinical setting remains a future step.