Watch the 70-second demo
Currently promoted
Decision Wall: ask 200 small questions at once
A way of thinking about AI judgments: instead of asking one big question, ask many small ones at once and let code decide. I built Decision Wall as an independent, open-source experiment in that idea, using TypeSafe's Jev model, which is made for answering many yes/no questions in one fast call. Type anything and 200 answers light up as a heatmap in about a quarter second. Seven question sets cover everything from scam texts to server logs to story pitches, and an Explain view shows which sentence drives each answer. Try it in the browser with recorded answers, or bring your own key. (Not affiliated with TypeSafe.)
From the Deckplates to the Data Center
My journey began troubleshooting radar systems on the USS Lake Champlain. That discipline—understanding the physics of the machine—now drives my work architecting petabyte-scale cloud infrastructure for the University of California.
I don't just write code; I design systems that build themselves. As the pioneer of Directed Agentic Engineering (DAE), I direct squads of AI agents to solve enterprise-scale problems.
In between came two decades in the machine room: running research clusters at Corning Incorporated, serving as an XSEDE and ACCESS Campus Champion at UCR, and serving as Vice Chair and then Chair of the UC Research IT Committee across the University of California system. Curriculum vitae
Director
Research Computing @ UCR
Architect
Agentic DevOps Systems
Educator
AI + HPC Field Manual
Past Chair
UC Research IT Committee
Building with AI since GPT-2
- 2019 TensorFlow and PyTorch GPU pods for UCR researchers on the National Research Platform. Repo
- 2020 GPT-2 jobs on NVIDIA RTX 8000 GPUs in the Nautilus cluster; deep learning notebooks for researchers.
- 2022 Stable Diffusion and Disco Diffusion notebooks for researchers, weeks before ChatGPT launched. Repo
- 2023 Ursa Major Ask, a GPT-3.5 assistant that writes and runs research code and Slurm scripts. Repo
- 2024 uma, an LLM in the Linux shell for HPC users. Repo
- 2025 Gemini multimodal video analysis, multi-agent simulations, generative media tools and the deep-research engine.
- 2026 MCP servers for HPC, research systems and the National Research Platform; a governed AI gateway; Directed Agentic Engineering.
The Vision: Directed Agentic Engineering
The future of engineering isn't writing code by hand. It's orchestrating autonomous squads.
My flagship system, Squad Manager, proves this. It is a recursive AI tool that deploys Architects, Engineers, and Auditors to build software from a single prompt. It built itself.
Flagship Innovations
ursa-bifrost: AI Meets the HPC Cluster
Let any AI assistant work a Slurm cluster, safely. One Go binary that is both an MCP server and a CLI. Claude, Gemini, Codex and Hermes get typed, audited, read-mostly access to jobs, failure diagnosis, costs and software, with Google sign-in, tiered permissions and two-step approval for anything that spends money.
A Family of MCP Servers for Research
AI assistants, connected safely to real systems. Beyond ursa-bifrost: nrp-mcp runs researchers' work on the National Research Platform's Nautilus GPU cluster from plain language, with NRP policy built in and nothing running until the person says yes. Others I have built stay private; each one has sign-in, roles and an audit trail.
Ursa Major HPC Cluster on Google Cloud
A cluster that costs nothing when idle. An elastic Slurm cluster built with Google Cloud's Cluster Toolkit: CPU, high-memory and NVIDIA L4 GPU partitions that grow on demand and shut down when the queue empties, a curated scientific software stack, per-project limits, and ursa-bifrost so AI assistants can submit and diagnose jobs.
Internal serviceGoverned AI Gateway
One governed door to frontier AI. A single OpenAI- and Anthropic-compatible endpoint to Gemini, Claude, open-weight and National Research Platform models, with per-lab budgets, spend limits and a full audit log.
Internal serviceNexus: The Research Knowledge Graph
The system of record for Research Computing. A Postgres and pgvector graph of researchers, labs, cloud projects, funding and every interaction, run through a CLI and an OAuth MCP server with role-gated tools, so AI assistants can work the graph with a full audit trail.
Private ArchitectureUltra AI Workstation
The AI operator desk. Email, calendar, tickets, Slack, the Nexus ledger and the HPC cluster in one keyboard-driven screen, with AI drafting under two-step approval and an agent that answers questions about anything on screen.
Squad Manager
The Autonomous Agency. A recursive Gemini skill that deploys specialized AI squads to build, test, and ship software autonomously.
View RepoDeep Research
The AI research engine. A CLI and web workstation for Gemini Deep Research: cited reports, notebooks, briefs and audio, plus a Lab that tests a report's claims as real jobs on the HPC cluster.
The Scientific Nexus
UCR HPC Job Gallery. An automated infrastructure engine that transforms production-ready Slurm scripts into a searchable knowledge base. Featuring 49 tested Slurm workflows across seven colleges and schools, from AlphaFold in biochemistry to engineering, business and the social sciences, with the index and gallery generated from the scripts themselves.
Explore the Job GallerySkywalker
The Fleet Auditor. An automated compliance engine for Google Cloud, enforcing "Ursa Major" security standards across the fleet.
UCR HPCC, 2017-2019
Where I started at UCR. As lead HPC administrator in the High Performance Computing Center I built the HPCC Cloud Service, which launched lab-private HPC clusters in AWS from the campus cluster with hpcc_cloud, a cfnCluster wrapper I wrote. Also 57 environment modules for the cluster's software stack and 19 workshops for researchers.
Applied Research & Experiments
Discover my latest projects that explore cutting-edge computing solutions, AI research, and data-driven innovations hosted in my experimental sandbox.
Research Computing Portal
A user-friendly management interface for HPC resources, researchers, and projects.
View ProjectAuto-Galaga Enhanced
An AI-driven version of the classic Galaga shooter featuring automated gameplay and top score tracking.
Play NowHPC Research Network
An interactive network map showcasing HPC connectivity and resources.
Explore MapDungeon Hero
A reinforcement learning environment set in a procedurally generated dungeon.
Try It OutText Encryptor & Decryptor
A simple web-based tool for encrypting and decrypting text using Caesar cipher logic.
Try It OutTrailblazer Maps
An interactive map with hiking trails, user location, and custom campsite markers.
Explore MapHAM Callsign Lookup
Quickly find detailed amateur radio callsign information using a clean, dark-themed interface.
View ToolLoRaWAN Visualization
Interactively explore LoRaWAN sensor and gateway RF coverage on a dynamic geographic map.
View ToolMJ Plant Simulation
A comprehensive research report and simulation guide covering soil, light, and biochemistry.
View ProjectHPC Workflow Simulator
Model, analyze, and optimize the execution of complex scientific applications on parallel compute systems.
View ProjectUCR Research Dashboard
Management interface for UCR Research Computing, HPC clusters, and Google Cloud operations.
View DashboardResearcher AI Prompts
A curated gallery of engineered AI prompts designed to assist researchers with complex data tasks.
View GalleryAI-Generated Adventure
A multimedia adventure page about hiking and kayaking, 100% autonomously generated by AI.
View PageWorkflow Builder
A drag-and-drop visual node tool to architect, connect, and run custom data pipelines.
Try It OutThe AI + HPC Field Manual
AI and high-performance computing are now one discipline. A 2026 field manual for the whole stack: GPUs and their memory, the fabrics that join them, the facilities that power them, and the training, inference and science that run on top.
Videos
Short narrated tours of things I've built. Each one is the real software, recorded live.