Lead Software Engineer, Agentic AI Systems

At Collective Health, we’re transforming how employers and their people engage with their health benefits by seamlessly integrating cutting-edge technology, compassionate service, and world-class user experience design. The Claims AI Automation team is currently evolving from traditional rule-based adjudication to an intelligence-driven platform. We are seeking a Lead Software Engineer who excels at high-level technical execution. This role is focused on the development and delivery of our Agentic AI strategy, turning architectural blueprints into production-ready systems. The ideal candidate will be the primary engine for building Python-based AI agents using the Google Cloud (GCP) ecosystem to automate complex claims workflows. You will be responsible for implementing sophisticated LLM behaviors using Gemini. What you will do: Execute Agentic Workflows: Build and deploy sophisticated AI agents using Google Vertex AI and the Google Agent SDK (ADK) based on provided architectural specifications. Develop and optimize agent behaviors using Gemini (Pro/Flash) with a focus on reliable tool-calling and multi-step reasoning; implement RAG and grounding strategies to ensure AI agents provide factual, data-driven responses derived from internal claims databases and policy documents. Design and implement complex system instructions, few-shot prompting, and Chain-of-Thought reasoning to ensure agents handle claims logic with high precision, performing Supervised Fine-Tuning on Gemini models to improve domain-specific performance in adjudication and medical coding Data & Messaging: Expert in SQL/PostgreSQ/AlloyDB/BigQueryUnderstand architectural decisions and actively drive reusable patterns for cloud-native, AI-enabled backend systems. Design and implement Stateful Orchestration patterns that incorporate human review breakpoints, ensuring agents can pause, persist state, and resume workflows after auditor approval. Implement Input/Output Guardrails and 'Circuit Breaker' logic to detect and intercept non-compliant agent behavior or toxic/hallucinated outputs in real-time before they reach production databases. Ability to implement Google Cloud DLP or similar tools within AI pipelines to ensure PHI (Patient Health Information) is never exposed to the model's training or logs. To be successful in this role, you'll need: Expert Python Developer : Exceptional skills in building production-grade applications using frameworks like LangChain, LangGraph, or PydanticAI . Gemini & Vertex AI : Hands-on experience building with the Gemini model family and the Google Agent SDK (ADK). Vertex AI Mastery: Deep experience with the Vertex AI Python SDK , specifically Reasoning Engine (Runtime) , Extensions , and Function Calling . Prompt Engineering : Proven ability to engineer high-performance prompts that mitigate hallucinations and ensure deterministic outputs in regulated environments. 8+ years of Full-Stack experience with a heavy focus on backend systems. AI-First SDLC : Mastery of AI-enhanced development tools (e.g., Cursor, WindSurf, GitHub Copilot ) to drive team velocity. Preferred Qualifications Healthcare Data: Familiarity with healthcare interoperability standards and claims-related data structures. Fine-Tuning Expertise: Practical experience in fine-tuning LLMs for specific industry nomenclature or structured output formats. Java Proficiency: working knowledge of Java and Spring Cloud, capable of supporting and extending existing microservice architectures. Our Tech Stack AI & LLMs: Gemini, Vertex AI, Google ADK, Python, LangChain. Existing Backend: Java, Spring Cloud, Spring Boot. Data & Messaging: PostgreSQL, Spark, AlloyDB, BigQuery Infrastructure: GCP (Vertex AI, GKE, Cloud Run), Docker. Pay Transparency Statement This is a hybrid position based out of our Lehi or Plano offices , with the expectation of being in office at least two weekdays per week. #LI-hybrid #LI-NM1 The actual pay rate offered within the range will depend on factors i