End-to-End Development of an AI-Driven Clinical Documentation and Data Processing Platform

Clinical documentation is a major time sink in Spanish elderly residential care, yet the voice-AI tools that could relieve it have been built almost entirely around English-speaking hospital medicine. This thesis, carried out during an internship at Vounded (Inevitable Software), addresses that gap on two fronts. It first develops a complete voice-to-note platform for Spanish care homes, built end to end: a professional records an encounter with one gesture, a speech-to-text stage transcribes it, a large language model returns a structured clinical note with its care controls and referrals extracted, and, once reviewed and signed, the note is written back into the facility’s management software, with all health data kept inside the European Union. It then evaluates the transcription layer rigorously, building a purpose-made Spanish medical-audio benchmark, the first of its kind, to compare nine commercial models on accuracy, speed, cost and noise robustness. AssemblyAI’s Universal-3 Pro led on accuracy (3.97% word error rate) and was the most noise-robust.