Senior Applied AI Engineer — Healthcare Claims Automation

We are looking for a senior applied AI engineer to build production AI automation for a healthcare SaaS platform serving emergency medical service organizations. This is not a chatbot, experimental prototype, prompt-writing, or foundation-model research role. You will build dependable AI-assisted workflows that operate in the background of a live claims and billing platform. Our long-term objective is to use AI throughout the claims lifecycle, including: - Documentation and claim-readiness review - Missing-information detection - Patient and insurance information discovery - Eligibility workflow assistance - Coding and modifier recommendations - Charge and payer-rule validation - Claim quality assurance - Denial classification and root-cause analysis - Appeal preparation - Work prioritization and intelligent routing - Operational summaries and anomaly detection AI will assist our professional billing team—not make uncontrolled clinical, coding, financial, or submission decisions. High-impact actions must include evidence, confidence, auditability, and appropriate human approval. OUR ENVIRONMENT Our existing platform includes: - TypeScript and Node.js - PostgreSQL on AWS RDS - React and React Native - AWS ECS, S3, SQS, and CloudWatch - Docker and GitHub Actions - Healthcare claims, clinical documentation, payer information, and financial workflows You will collaborate with the product owner, a senior backend/database engineer, a frontend engineer, billing professionals, and AI development tools. WHAT YOU WILL OWN - Architecture for background AI automation - LLM workflow and agent orchestration - Structured outputs and tool/function calling - Retrieval grounded in approved source material - Document extraction and classification - Model and provider selection - Prompt and workflow versioning - Evaluation datasets and automated regression testing - Confidence scoring and escalation thresholds - Human-review and approval workflows - Output provenance, citations, and audit trails - Model quality, latency, reliability, and cost monitoring - Safe retries, idempotency, queues, and failure handling - Protection of PHI and other sensitive information - Preventing sensitive data from appearing in logs or development tools - Documentation of system behavior, limitations, and risks THE FIRST PROJECT Your initial responsibility will be designing the foundation for claims AI automation. This will include: 1. Mapping candidate claims workflows by impact, risk, and feasibility. 2. Separating deterministic business rules from tasks that genuinely benefit from AI. 3. Defining the source information available to each workflow. 4. Creating a reusable AI execution architecture with structured inputs and outputs. 5. Establishing evaluation datasets using approved, protected examples. 6. Building automated quality, safety, and regression tests. 7. Implementing confidence thresholds and human-review requirements. 8. Creating complete audit records showing what information was used, which model and workflow version ran, and why an output was produced. 9. Monitoring accuracy, latency, failures, and cost. 10. Delivering one narrow production workflow before expanding into additional automation. We want small, measurable, dependable systems—not an unsupervised “AI agent” with broad database access. REQUIRED EXPERIENCE - Production experience building LLM-powered applications - Strong Python and/or TypeScript/Node.js engineering - OpenAI API, AWS Bedrock, Anthropic, or comparable model APIs - Structured output, tool calling, retrieval, and workflow orchestration - Evaluation datasets and automated AI regression testing - PostgreSQL, APIs, queues, and background jobs - Reliable handling of retries, timeouts, and partial failures - AI observability, tracing, token usage, latency, and cost monitoring - Security and privacy for sensitive information - Human-in-the-loop workflow design - Strong written and spoken English - At least 2–3 hours of overlap with US Eastern Time STRONG ADVANTAGES - Healthcare claims, billing, revenue-cycle, or insurance experience - Ambulance or emergency medical services billing - X12 837, 835, 270/271, or 276/277 transactions - Claim denials and appeals - Medical coding workflows - OCR and document intelligence - pgvector, embeddings, and retrieval systems - HIPAA-regulated production systems - Experience comparing multiple models and providers - Experience turning prototypes into monitored production systems HOW WE WORK The product owner already works extensively with AI tools. We are not hiring someone merely to use ChatGPT, write prompts, or generate ordinary application code. We need someone who can: - Decide where AI should and should not be used - Design measurable automation with clear success criteria - Build reliable production systems around nondeterministic models - Identify hallucinations, data leakage, and unsafe automation risks - Evaluate outputs against real expert-reviewed examples - Work effectively with billing subject-matter experts - Explain technical tradeoffs to a non-technical product owner - Challenge ideas that cannot be implemented safely or measured properly ENGAGEMENT DETAILS - Long-term hourly engagement - Approximately 15–20 hours per week initially - Opportunity to expand as successful automations enter production - Paid, clearly scoped trial project - Individual freelancers preferred - You must personally perform the work - No undisclosed subcontractors or shared access TO APPLY Please answer every question: 1. Describe a production LLM system you personally built. What did it automate, and how did you measure its accuracy? 2. How would you evaluate an AI system that reviews healthcare claims before submission? 3. How do you prevent hallucinated facts or recommendations from entering a production workflow? 4. How would you decide which claims tasks should use deterministic rules, conventional software, or an LLM? 5. Describe how you implement structured outputs, retries, idempotency, human review, and audit trails. 6. How have you protected sensitive or regulated information when using external AI providers? 7. What experience do you have with healthcare claims, billing, denials, insurance, or medical documentation? 8. What hours can you reliably overlap with US Eastern Time? 9. Will you personally perform all work and be the only person accessing our code and systems? Generic proposals that do not answer these questions will not be considered.