Skip to content

Ben Vierck

About#

Ben Vierck

Founder of Positronic AI — enterprise AI, deep learning, and LLMs. Former VP of Engineering at Fujitsu; began on Microsoft's Windows team.

Email · LinkedIn · GitHub · Essays

Now#

Building lit.ai, an enterprise AI platform that runs entirely on the customer's own infrastructure. Fractional CTO for two companies. Writing an essay series on where the AI windfall goes; the latest is Where the Nickel Went. On weekends, building Sail, a sailing game with no engines, where reading the water and trimming the sheet is the whole skill.

Essays#

I write at essays.xcud.com: essays from a practitioner on what the machine does to work, prices, and the people in between. Start with Where the Nickel Went, The Last Wide Rung, or The Cost of Software Is Now Zero. Subscribe to get new ones by email.

Bio#

My career began on Microsoft's Windows Build team [3] in the late 1990s, compiling an operating system that would run on hundreds of millions of desktops. In the thirty years since, I've alternated between leading engineering inside large companies and founding my own. Each has sharpened the other.

As Vice President of Engineering at a Fujitsu subsidiary [1], I led a hundred-person organization, held a seat on the executive team reporting to the CEO, and shepherded the company through its move from on-premise systems to the cloud.

I founded Positronic AI and grew it, without outside capital, to a million dollars a year in deep-learning work that was never academic: diagnosing disease in EEGs, catching counterfeit coins, predicting from video the movements that precede an accident. Five patent applications [4] came out of those years. One of the clients — the owner of the EEG company — thought enough of the work to become an investor, putting in a million dollars and taking over sales.

That investment built lit.ai, an enterprise AI platform that runs entirely on a customer's own infrastructure. Out of building it came one more patent: a way to use a language model to tune a neural network's training in real time. Published as US 2026/0105319 A1, April 16, 2026.

Software#

  • Positronic lit.ai: a platform for building deep learning models on your own hardware.
  • Fujitsu PaperStream Capture Server: web-based document capture, classification, and extraction.
  • Devfarm PowerWF, PowerSE, PowerVI: PowerShell script authoring and management tools.
  • Quest Benchmark Factory, Foglight Transaction Recorder: database and web server load testing. [2]
  • Microsoft Windows: build team, late 1990s.

Patents#

Positronic AI — real-time hyperparameter tuning of neural-network training via language models
U.S. Patent Application No. 19/360,775 (provisional 63/708,059), published as US 2026/0105319 A1, April 16, 2026

Clinical EEG deep learning — co-authored for Stratus / Merln

A body of five patent applications introducing novel techniques for training artificial neural networks on EEG signal data: hot-swapping irregular segments, probabilistic sample weighting, pooled frequency downsampling, adaptive feature clipping, and adaptive normalized PSD. Three matured to published U.S. applications:

US-20240112023-A1 — Preserving signals in data inputs with moderate-to-high variances in data sequence lengths for artificial neural network model training
US-20230177325-A1 — Preserving signals for artificial neural networks when downsampling
US-20230316072-A1 — Artificial neural network for sparse data processing

Talks#

Software Engineering Nankai University Guest Lecture

This personal perspective on the history of Software Engineering follows the state of Software Engineering from its birth in 1968 through the present. The content of this talk was heavily influenced by Paulo Perrota's talk at the 2012 Ruby Conference.

Publications and Articles#

  • VMware vSphere PowerCLI Reference: Automating vSphere Administration by Luc Dekens (First Edition, Chapter 22)
  • Essential Blogging published by O'Reilly (Acknowledgement)
  • Artificial Intelligence for Humans by Jeff Heaton (Acknowledgement)
  • Microsoft Scripting Blog Articles:
    • "PSImaging Part 1: Test-Image" - March 19, 2015
    • "PowerTip: Read .exe Magic Number with PowerShell" - March 19, 2015
    • "PSImaging Part 2: Export-Text from Images" - March 20, 2015
    • "PSImaging Part 3: Group-ImageFile" - March 21, 2015

Conferences and Events#

  • FIRST Robotics Competition: Deep Space, 2019 (Judge & Speaker)
  • StartupGrind Global 2018 (Top 20 Finalist, Top 50 Startup)
  • StampedeCon 2017 (Speaker)
  • VMworld 2009 (Speaker)

Footnotes#

[1]: The subsidiary was KnowledgeLake, wholly owned by Fujitsu while I was there. I ran a quality initiative across the product line, retired the products that weren't earning their keep and put the money into the ones that were, and charted the move from our on-premise platform to the cloud. I tried, and failed, to get Fujitsu to invest in machine learning. I reported to the CEO and sat on the executive team.
[2]: At Quest Software I was a development manager. I architected Benchmark Factory for Web Servers, a load-testing tool with the industry's first browser record-and-playback, and grew its team to twelve. I also ran development of the Management Packs that let System Center manage Unix hosts, and I was the one technical member of the executive mergers-and-acquisitions team.
[3]: I was hired onto the Windows Build team in 1996. Volunteering for kernel bug triage got me into projects like Arbiter Weekend, DLL Hell, and Hydra, and I learned the trade from some of the best software engineers in the world.