Job Description

What You Will Do

• Design, implement, and operate multi-agent systems using frameworks such as LangGraph, LangChain, or similar, to solve complex, non-linear business workflows

• Implement advanced reasoning patterns (Chain-of-Thought, ReAct, Plan-and-Execute, tool-use) so agents can decompose high-level goals into executable plans and coordinate safely with tools, services, and data sources.

• Build robust agentic loops using Python AsyncIO, including concurrency control, timeout handling, and graceful degradation strategies for long-running or multi-step tasks.

• Architect and optimize end-to-end RAG pipelines, including document ingestion, chunking strategies tuned to insurance documents, hybrid search (semantic + keyword), and retrieval-time re-ranking.

• Implement query expansion, result fusion, and personalization strategies to maximize answer accuracy and reliability for production use cases.

• Use the Model Context Protocol (MCP) or similar standards to define and implement connectors between agents and Guidewire’s data sources, APIs, knowledge bases, and external tools.

• Optimize for token usage, latency, and throughput while preserving reasoning quality, applying strategies such as caching, retrieval pre-filtering, dynamic model routing, and hierarchical prompting.

• Uphold and model Guidewire’s culture of determination, collaboration, continuous improvement, and bravery in how you design, build, and ship AI systems.

What We’re Looking For

• Strong software engineering background (typically 8+ years) with production experience in building distributed systems, APIs, or data-intensive applications.

• Expert-level Python skills, including AsyncIO for concurrent and high-throughput workflows, plus familiarity with TypeScript/Node.js for building modern AI middleware and integration services.

• Hands-on experience with at least one major AI orchestration framework (e.g., LangChain, LangGraph, LlamaIndex, AutoGen, CrewAI) and building multi-step or tool-using agents in production.

• Proven experience designing and operating RAG pipelines: vector stores, embeddings, chunking strategies, hybrid search (semantic + lexical), and re-ranking or scoring strategies.

• Experience with at least one vector database or search engine (e.g., OpenSearch, Elasticsearch, Pinecone, Weaviate, PGVector, or similar) and one relational or NoSQL database.

• Production experience integrating with one or more LLM APIs (OpenAI, Anthropic, Gemini) and working with open-source models (Llama 3+, Mistral, etc.), including evaluation and prompt engineering.

• Solid understanding of cloud-native development (Docker, Kubernetes) and at least one major cloud platform (AWS, Azure, or GCP), ideally including AI-specific services (e.g., Bedrock).