Yesi Xie

Senior Software Engineer with 12 years of experience designing and delivering production-grade agentic AI systems, LLM pipelines, and enterprise knowledge retrieval architectures. Builds end-to-end solutions across LangGraph, LangChain, AWS Bedrock, Azure OpenAI, Neo4j, and Pinecone for regulated enterprise environments spanning fintech, healthcare, and e-commerce. Brings a strong consulting and cross-functional delivery background, owning architecture decisions from discovery workshops through production deployment with clear communication to VP and CXO-level stakeholders.

• Architected a multi-agent orchestration framework using LangGraph and LangChain for a fintech document intelligence platform, designing planning, execution, memory, and tool-use layers across 3 autonomous agent types handling regulatory compliance checks and financial reporting workflows
• Designed Graph RAG pipelines on Neo4j to power enterprise knowledge retrieval across structured fintech datasets, enabling relationship-aware context injection into LLM prompts with sub-second retrieval latency
• Defined prompt versioning and evaluation frameworks on Azure OpenAI, establishing model monitoring guardrails and FinOps controls to manage token cost and output reliability across 40+ agentic endpoints
• Built Vector RAG pipelines using Pinecone and LangChain for a financial document Q&A system, chunking and embedding policy and audit documents to support SOX-compliant retrieval workflows
• Introduced human-in-the-loop checkpoints and structured audit trails into agentic workflows, ensuring GDPR and SOX compliance requirements were enforced at each approval and escalation boundary
• Ran discovery workshops with business and technical stakeholders across 2 fintech product lines, extracting requirements and producing scoped solution architecture documents reviewed by VP-level audiences
• Presented architecture trade-offs, risk assessments, and LLM platform recommendations to CXO-level audiences, translating technical complexity into clear proposal-ready decision frameworks
• Designed MCP integration patterns connecting LLM agents to enterprise Salesforce and ServiceNow data sources, enabling cross-system tool use within agentic task pipelines
• Defined A2A integration patterns for agent-to-agent communication across distributed orchestration layers, reducing inter-agent coordination latency and increasing workflow reliability
• Deployed agentic platform infrastructure on AWS Bedrock, selecting foundation models per use case and integrating with existing enterprise data systems through structured API adapters
• Established model evaluation pipelines with automated regression benchmarks across 3 major LLM release cycles, catching prompt drift and output degradation before production rollout
• Mentored 3 engineers on agentic AI design principles, Graph DB schema modeling, and LLMOps practices, raising the team's architectural fluency across the AI stack
• Owned end-to-end delivery of 2 production agentic systems in regulated fintech environments, managing estimation frameworks, assumptions logs, and stakeholder sign-off from scoping through deployment

• Master in Computer Science - University of Pittsburgh
• Bachelor in Computer Science - Wuhan Institute of Technology