Propel Assistant — Healthcare AI Communication Platform
Voice-activated healthcare AI assistant with multilingual support, enabling real-time patient-provider communication across language barriers.
40+ language support including medical terminology translation
Sub-3-second latency for real-time conversational flow
95% accuracy on medical terminology translation (validated by certified translators)
Deployed in pilot program at 3 community health centers
Healthcare providers in multilingual communities face critical communication breakdowns with non-English-speaking patients, leading to misdiagnoses, medication errors, and reduced treatment adherence. Traditional translation services add 10-20 minutes per consultation and cost $100-200 per session, making them economically unfeasible for routine appointments in underserved clinics.
We built a voice-activated AI assistant that provides real-time medical translation, symptom documentation, and care plan explanations in 40+ languages. The system uses OpenAI's Whisper for speech recognition and GPT-4 for context-aware medical translation, ensuring clinical accuracy while maintaining conversational flow.
The Challenge
The United States healthcare system serves increasingly diverse populations, yet language barriers remain a critical patient safety issue. The Joint Commission estimates that language discordance between patients and providers contributes to tens of thousands of adverse events annually, including medication errors, surgical complications, and missed diagnoses.
Community health centers and safety-net hospitals serve predominantly immigrant populations but operate on razor-thin margins that can't absorb $100-200 per session for professional medical interpreters. Phone and video translation services introduce 10-20 minute delays and disrupt the patient-provider relationship, leading clinicians to avoid them except in emergencies.
Existing consumer translation apps (Google Translate, iTranslate) fail in medical contexts—they lack domain-specific terminology, can't handle complex symptom descriptions, and introduce dangerous ambiguities. A mistranslation of "chest pain" versus "chest pressure" could mean the difference between appropriate cardiac workup and a missed heart attack.
The result is a silent crisis: non-English-speaking patients receive lower-quality care, experience worse health outcomes, and are less likely to follow treatment plans they don't fully understand.
Our Approach
We designed Propel Assistant as a clinically-aware translation layer that preserves the human patient-provider relationship:
-
Voice-First Interface - Hands-free operation via wake word activation ("Hey Propel") allows providers to maintain eye contact and physical examination focus. Whisper speech recognition achieves 97% accuracy on medical terminology across English, Spanish, Mandarin, Arabic, and Vietnamese.
-
Medical Context Engine - Fine-tuned GPT-4 model trained on 50,000+ clinical encounters ensures translations preserve critical medical nuances. The system distinguishes between "dizzy" (vertigo) and "lightheaded" (presyncope), flags dangerous symptom combinations, and suggests follow-up questions providers might miss.
-
Bidirectional Real-Time Translation - Sub-3-second latency enables natural conversation flow. Provider speaks in English, patient hears translation via smartphone speaker in their language; patient responds, provider receives English translation via wireless earpiece.
-
Automated Documentation - Generates structured SOAP notes (Subjective, Objective, Assessment, Plan) from conversation transcripts, reducing post-visit charting burden by 40%. Integrates with Epic and Cerner EHR systems via FHIR API.
Results & Impact
Deployed in pilot programs at three community health centers in Columbus, Los Angeles, and Houston—serving predominantly Spanish, Mandarin, and Arabic-speaking populations. Early results demonstrate both clinical and operational impact:
Medical accuracy validation by certified medical interpreters showed 95% fidelity on symptom translation and 98% on medication instructions. The 5% error rate consisted entirely of minor phrasing differences, not clinical misinterpretations.
Average consultation time decreased by 8 minutes compared to phone interpreter services, allowing providers to see 2-3 additional patients per day. Clinics reported 15% revenue increase from improved visit throughput, making the $200/month per-provider subscription cost effectively free.
Patient satisfaction scores (measured via post-visit surveys) increased from 67% to 89% for non-English-speaking patients. Qualitative feedback highlighted "feeling heard" and "understanding my treatment" as primary factors. Follow-up appointment attendance improved from 61% to 78%, suggesting better comprehension of care plans.
The automated documentation feature proved unexpectedly valuable: providers appreciated having verbatim patient quotes for medical-legal protection, and coding specialists used detailed symptom descriptions to support higher-acuity billing codes where appropriate.
Propel Assistant demonstrates how AI can address healthcare equity not through radical disruption, but by removing friction from essential human interactions—letting doctors be doctors and patients be understood.