Senior Software Engineer - Applied AI

What You’ll Do

Your primary focus is agent design and orchestration. Multi-agent orchestration layer. Design and build the LangGraph-based system that classifies member intent and routes to the right agent. Stateful, context-preserving, failure-tolerant. Handles commercial agents and internally built capabilities through the same interface.

Agent integration contract. The standardized pattern that makes any agent pluggable: input schema, output schema, confidence signaling, latency SLA, fallback behavior.

Context management across turns. Multi-turn conversations that maintain full member context across agent handoffs.

Guardrail and safety layer. Clinical safety boundaries, PHI handling at the LLM layer, and response normalization ensuring every agent response arrives in a consistent tone and within compliance constraints regardless of which agent generated it.

RAG pipeline architecture. Benefits documents, formularies, and policy content retrieval via Azure AI Search.

You define the ingestion strategy, chunking approach, hybrid search configuration, and retrieval quality standards.

Prompt architecture. Versioned prompt registry, structured output contracts, few-shot pattern design. Every prompt is sable, rollback-able, and understandable by the team.

To Be Considered For This Position

• You must have a Bachelor’s degree and 4-7 years of experience in software engineering.
• You must have experience with agentic AI in production—not LLM wrappers, not single-agent chatbot demos, not RAG tutorials. Systems with multiple agents, explicit state management, and real users depending on the answers being correct.
• You must have 2–3 years building and operating production LLM or agentic AI systems.
• You can describe specific decisions you made in a multi-agent system: state management approach, how you handled agent failures mid-conversation, how you kept the system governable as agents were added.
• You must know where it breaks down and how you worked around it.
• You must have experience debugging retrieval quality problems, iterated on chunking strategies, dealt with stale knowledge bases in a live system with real users.
• You must have built evaluation systems from scratch: golden datasets, automated regression, domain-appropriate accuracy metrics. Your eval suite caught a production regression before users reported it.
• You must have experience in healthcare, insurance, fintech, or legal domains—AI systems where a wrong answer has consequences, and you designed the safeguards that make accuracy non-negotiable.
• You must have strong knowledge of Python. Work in an existing codebase and extend it with the AI layer.
• The target stack is Python / FastAPI, LangGraph, Azure OpenAI / Claude via Azure, Azure AI Search, Azure Container Apps.