Vera: Voice Follow-up After Surgery, in Spanish

After surgery, patients go home with a discharge plan on paper, and the complications that matter (an infected wound, a fever, chest pain) show up days later, at home. Follow-up is a nurse on the phone, and no clinic has the hours to call every patient. Vera is a voice agent that makes those calls in the Spanish patients actually speak in Colombia ("calentura", "me siento maluca", "me dio un yeyo"). It follows a nurse's agenda (pain, wound, fever, eating), answers questions only from the patient's own discharge plan and public-domain clinical guidance, and shows which document backs each answer. A question the documents can't answer never reaches the model. Its core design choice: the model is never the gatekeeper of its own supervision. Vera drives AssemblyAI Universal-Streaming directly, so a deterministic engine with a Colombian clinical lexicon reads every partial transcript before any model. A red flag fires a median of 0.85 s before the patient finishes the sentence; the emergency reply is written by code and pre-synthesized, and the alert is posted to the care team's Slack in English. Claude, through the AssemblyAI LLM Gateway, adds a second safety layer as a risk judge, and Cartesia gives Vera a native Colombian voice. Built for hospitals and clinics that operate: USD 4 per patient followed, about USD 0.07 per call in measured cost. Tested with twelve whole calls against the real model. Watch the alerts arrive: https://join.slack.com/t/veracareteam/shared_invite/zt-4b18ojlw7-bQxL2hOA3rdD8CRh_V8rvg