Senior AI Engineer

MD Ally builds telehealth solutions that connect eligible non-emergency 911 callers with clinicians. The Senior AI Engineer will architect and scale production AI systems for clinician productivity, patient care navigation, and clinical workflows, with a focus on safety, privacy, evaluation, and observability.

Responsibilities:

• Build audio-to-text pipelines using models like Whisper or Deepgram combined with LLMs that listen to telehealth video visits, filter out small talk, and draft structured SOAP notes (Subjective, Objective, Assessment, Plan)
• Build integrations using FHIR or HL7 standards that pull a patient's past labs and medications before a visit, and push approved clinical notes, ICD-10 billing codes, or draft prescriptions back into systems like Epic, Athenahealth, or Elation
• Architect Retrieval-Augmented Generation (RAG) systems that digest dozens of pages of messy PDF medical records or specialist referrals into clean, concise summaries with exact source citations for the provider
• Build conversational agents that gather a patient's medical history and symptoms before a video call, and ask follow-up questions based on clinical protocols
• Develop classification systems that analyze patient messages to route urgent red-flag symptoms (such as chest pain) immediately to an on-call nurse, while routing routine prescription refills or scheduling requests to automated workflows
• Create automated systems that translate complex medical jargon from a doctor's note into plain-language instructions, medication reminders, and check-ins in the patient's preferred language
• Build middleware that strips names, dates of birth, and SSNs before data reaches external APIs, routing all traffic strictly through HIPAA-compliant endpoints (such as AWS Bedrock, Azure OpenAI, or GCP Vertex AI) under a Business Associate Agreement (BAA)
• Implement real-time validation layers that block models from diagnosing conditions, recommending unapproved treatments, or hallucinating drug dosages, forcing standard fallback responses whenever clinical boundaries are crossed
• Engineer low-latency streaming infrastructure using WebSockets and WebRTC so voice agents and live transcription tools respond in milliseconds during live telehealth calls without awkward pauses
• Build AI agents programmed to act like patients with specific conditions, accents, or edge-case behaviors to stress-test intake workflows before real patients interact with them
• Build automated test suites that grade generated SOAP notes against gold-standard physician notes, checking for omitted allergies, incorrect dosages, or hallucinated symptoms
• Maintain full logging systems so that if a clinician or regulator asks why the system suggested an output, our team can trace the exact prompt, retrieved medical document, and model version used