Brett Vogelsang

About Brett Vogelsang

I build enterprise multi-agent systems. At SAS, I\'ve spent the last year figuring out how to get LLMs to work with real analytical models, not just alongside them. Supply chain optimization models, demand forecasting engines… the kind of math that actually runs businesses. I design the orchestration (LangGraph, Google ADK), wire up the interop (A2A protocol), and build the full stack around it. TypeScript/React on the front, FastAPI and Python on the back. My focus is the intersection of AI agents and enterprise analytics. Getting LLMs to actually do useful work against complex systems is harder than most people realize. I\'ve spent time on the coordination layer, making sure context doesn\'t get lost across agent handoffs and that humans stay meaningfully in control when the stakes are high. Honestly though, none of that matters if the product falls apart at the UI layer. I know what it feels like when an agentic experience gets it wrong, and I\'ve spent time making sure ours didn\'t. I\'ve reviewed how agent state surfaces to users, shaped how approval flows work, and stepped in to build components myself when I needed to. I care as much about what happens at the interface as what\'s running underneath. If you\'re working on hard problems at the intersection of AI and enterprise software, let\'s connect. My stack: Python · LangGraph · LangChain · ADK · A2A · MCP · FastAPI · Docker · PostgreSQL · Opik · LangSmith · TypeScript · Jira · GitHub

Experience

1. Senior Associate Data Scientist

   SAS

   Aug 2024 — Present · Cary, NC, US

   Built LangGraph agents using a centralized orchestrator and subagent architecture. The supervisor routes requests to specialized subagents in parallel, each stateless and scoped to their own domain. Results synthesize back through the orchestrator into a single coherent response- Built an agent interoperability platform using hub-and-spoke architecture and Google\'s A2A protocol. Agents from different developers use A2A + our platform to discover each other at runtime, connect through a central orchestrator, and communicate without hard-coded dependencies between them- Built a context management system using progressive disclosure. Agents surface only the context relevant to each task, with automatic summarization before hitting model limits and semantic search to pull relevant history from past sessions- Designed model agnostic agent systems utilizing Azure AI Foundry. Used Opik to A/B test LLM performance across subagents on domain-specific prompts and toolsets, identifying the lowest-cost, highest-performing model for each task in the graph- Presented enterprise multi-agent architecture to C-suite leadership, making the technical case for a production AI agent platform at SAS with Supply Chain Agents as the pilot. Received executive buy-in on full product development and have since led milestone presentations to keep leadership aligned as the product evolves.

Education

• Lenoir-Rhyne University

  Bachelor of Science - BS, Computer Science

  2017 — 2021

• Institute for Advanced Analytics

  Master of Science - MS, Analytics

  2021 — 2022