Brad Merrill — Independent AI Researcher · Local LLM Deployment · Distributed Systems · GPU Infrastructure. Greater Seattle Area.

Distributed systems fail in unpredictable ways. Every phase of my career has been about solving that problem in a different domain — transaction processing, GPU computing, cloud microservices, and now local AI deployment. The abstractions shift. The core challenge doesn’t.

My current research focuses on the practical challenges of local LLM deployment — model pruning, retrieval-augmented generation (RAG), and building continuous deployment pipelines for AI systems. I’m actively working with SGLang for structured LLM generation, FAISS and cuVS for GPU-accelerated vector search, the NVIDIA NeMo Agent Toolkit for agentic workflows, and Open WebUI for local model serving. I’m also exploring small language model research and applying LLMs to structured reasoning tasks. The through-line is the same question I’ve been asking my whole career: how do you build systems that are reliable, efficient, and deployable at scale?

Running large models locally, efficiently, and reliably — with proper retrieval augmentation, sensible deployment pipelines, and the ability to update and prune models without disrupting production — is a distributed systems problem at its core. The tools have changed dramatically. The fundamental challenges haven’t.

Current work

I maintain builds of the open-source AI stack tuned for NVIDIA GB10 (DGX Spark) and RTX 40/50-series GPUs, each on a tuned-builds branch of my fork, with shared build tooling in tuned-common:

Career

  • 2024–present · ZBrad LLC — Independent AI Researcher. Local LLM deployment, model pruning, RAG pipelines, and continuous deployment for AI systems, including LLMs for chess move generation and evaluation.
  • 2016–2024 · NVIDIA — Senior Software Engineer. Cloud infrastructure across GeForce content pipelines, a GPU-as-a-service streaming platform for professional 3D/CAD applications, vGPU licensing and utilization, and Kubernetes hosting standards on AWS.
  • 2012–2015 · Microsoft — Senior Technical Program Manager / Senior SDE on Azure Service Fabric, shaping the developer model for cloud-native microservices with internal teams and early-adoption customers.
  • 2009–2012 · Clearwire, then Ericsson — Development and engineering manager for network operations systems: embedded Linux at microwave tower sites and network-element ETL into a data warehouse.
  • 2007–2009 · InfoSpace — Senior Development Manager for Dogpile.com meta-search.
  • 2006–2007 · CDigix — Director of Development: DRM-protected academic media and student music streaming.
  • 1996–2006 · Microsoft — Java and .NET evangelist (helping ISVs build compilers for the CLR), SDE on ASP.NET, and TPM on Dynamics.
  • 1993–1996 · Sybase — X/Open XA integration with TP monitors, including Tuxedo.
  • 1980–1993 · Digital Equipment Corporation — OS internals, compilers, and transaction processing: DECintact, DECdtm, and Reliable Transaction Router, deployed at NASDAQ, Bank of New York, the Swiss Stock Exchange and Indian Railways.

Publications

  • C# Essentials
  • C# Design Patterns and VB.Net Design Patterns

Recognition

  • Microsoft OneWeek Hackathon — First Place, Consumer Category

Find me on GitHub and LinkedIn.