24Seven: A Mobile AI Surgical Practice Workflow Assistant-Design, Implementation, Pilot Data

Background: Effective patient-surgeon communication, history taking, and document intake are foundational to surgical care, but typical preoperative workflows remain inefficient and variable. Existing health apps or chatbots often lack surgical specificity, strong safety guardrails, or structured outputs for clinician-use. Objective: To build, deploy, and pilot 24Seven, a mobile AI assistant designed to standardize pre-consultation history taking and document intake in a surgical practice, enforce safety boundaries, and generate structured clinician reports, while ensuring GDPR-compliance, usable patient experience while avoiding replacing human medical help. Methods: The assistant was built as a mobile-optimized web application via Base44, using a supervised low-code UI environment. Backend orchestrates prompt-engineered LLM calls, enforces policies: no prescribing, only differential outputs, red-flag detection, and secure file uploads. The structured interview covers eight domains; after completion (or idle timeout) a summary is emailed to the surgeon. Pilot deployment across 25 patients over 2 weeks collected metrics on completeness, safety compliance, report delay, patient satisfaction (Likert scale), and conversational coherence. Results: Pilot data (n=25) showed: Domain coverage: 99% of all domains filled; missing data in one patient due to network drop. Policy compliance: 0 instances of diagnosis or treatment advice. Differential outputs produced in 100% of cases where diagnosis was requested. Report delay: mean time 3.2 (±1.1) minutes post‐interview completion. Patient satisfaction: average 4.6/5 on clarity; 4.5/5 on tone; 4.2/5 on perceived usefulness. Red-flag escalations: triggered in 3 of 25 (12%) cases, appropriately. Conclusions: The pilot suggests 24Seven is feasible, safe under constrained policy, satisfactory to users, and efficient in producing clinician‐usable reports. It is currently used in our practice with warnings about trial period debugging.