Work with me
Work with me
I am Rowan Brad Quni-Gudzinas, and I build research systems that people can check. I have spent 15 years turning data into public decisions, including national research programmes at the U.S. Federal Highway Administration and at AARP's Public Policy Institute, where I led the Livability Index. Since 2024 I have run QNFO, an independent research imprint that asks what computation really costs and delivers: energy per correct answer (Joules-per-Solution), what an AI-assisted claim is worth, and what an autonomous research system actually delivers.
Energy per correct answer (JPCUB) assessment
Who it is for
Teams building a computing platform (quantum, AI inference or HPC) and data-centre operators who need an energy figure tied to correct results, not to peak throughput.
What you get
- A measurement plan under the published Joules-per-Solution protocol: the task, the check that decides whether an answer is correct, and a whole-system boundary that includes memory, I/O, cooling and power conversion, not only the processor.
- The measured joules per correct answer, with its uncertainty stated and the protocol's anti-gaming provisions (pre-registration, adversarial validation) applied.
- A written report, published only if you agree.
The protocol is published. The JPCUB figures published so far, including those for 17 quantum platforms, are estimates built from published specifications and third-party data, not metered measurements, so a first engagement is also the protocol's first field test, and the report says so.
How to start
Email me the system, the workload you care about, and the metering you already have (rack, facility or wall plug).
Based on: The Joules-per-Solution Metric: Definition, Measurement Protocol, and Anti-Gaming Provisions; JPCUB Competitive Landscape v2.0 (17 platforms); Joules-per-Solution for Stochastic and Agentic Inference (LLMs).
Email me about thisSubject starts with [work-with-me:jpcub]Review of an AI research or agent operation
Who it is for
Teams running AI agents or an AI-assisted research pipeline in production who want to know what it costs, what it delivers and what it gets wrong.
What you get
- A cost line: what the operation spends per month and per delivered result, from your bills and logs.
- A delivery line: what it actually ships, measured from its outputs rather than from the agents' own reports.
- A failure ledger: each recurring failure with its evidence, and the check or rule that would stop it coming back, ranked by what it costs you.
This is the method I use on my own system, a fleet of 44 deployed Cloudflare Workers (October 2026), whose objectives, successes and failures are published as a ledger.
How to start
Email me what the operation does, its agents and models, roughly what it spends a month, and which logs and bills you can share.
Based on: Operating the Quniverse Fleet: Objectives, Successes, Failures, Roadmap.
Email me about thisSubject starts with [work-with-me:agent-review]Talks and workshops
Who it is for
Conferences, labs, research offices and engineering teams.
Topics
- Joules per correct answer: what computation really costs, and how to measure it without gaming the number.
- Running an autonomous research system: what it delivered, what it cost and what went wrong, from a published failure ledger.
- Reading an AI-assisted claim: ignorance audits and epistemic legibility, as a talk or as a workshop that audits your own AI-assisted work.
Remote by default. Slides and materials are shared afterwards.
How to start
Email me the audience, the date and the format (talk, panel or workshop).
Based on: The Joules-per-Solution Metric: Definition, Measurement Protocol, and Anti-Gaming Provisions; The Universal Ignorance Audit; Epistemic Legibility in AI-Assisted Science; Operating the Quniverse Fleet: Objectives, Successes, Failures, Roadmap.
Email me about thisSubject starts with [work-with-me:talk]Research collaboration
Who it is for
Researchers in energy-aware computing, metascience and AI oversight.
Three open lines
- JPCUB measurements: run the protocol on hardware you operate, or test its anti-gaming provisions. Status: protocol published; first measurements wanted.
- The ignorance audit: apply the Universal Ignorance Audit to an AI-assisted corpus or pipeline and publish what it finds. Status: method published.
- The agent-correction dataset: an open, de-identified record of the corrections a running AI agent fleet receives (rule changes, blocked changes, owner overrides, reopened false closures) and what happened next. Status: planned; I am looking for researchers in AI oversight and corrigibility to shape it.
Results are published openly with DOIs, and every contributor is credited.
How to start
Email me which line interests you and what you would bring: hardware, data or a method.
Based on: The Joules-per-Solution Metric: Definition, Measurement Protocol, and Anti-Gaming Provisions; The Universal Ignorance Audit; Operating the Quniverse Fleet: Objectives, Successes, Failures, Roadmap.
Email me about thisSubject starts with [work-with-me:research]Roles in research management and applied AI
Who it is for
Organisations hiring for research management, applied AI, or data and policy research leadership.
What I bring
- Research programme management: a $1.5M federal research portfolio managed as a certified Contracting Officer's Representative at the U.S. Federal Highway Administration.
- National data products: I led the AARP Livability Index, which scores U.S. neighborhoods from 50+ data sources across 7 domains.
- AI systems in production: I built and run QNFO's autonomous research system and publish what it costs and where it fails.
Remote, based in Amsterdam; EU and US hours.
How to start
Email me the role and a link to its description. My CV is on Zenodo.
CV: doi:10.5281/zenodo.23082080. Earlier work is published under Brad Gudzinas.
Email me about thisSubject starts with [work-with-me:role]The record
- 2011 to 2015U.S. Federal Highway Administration, Data Analyst and Research Manager. Managed a $1.5M federal research portfolio as a certified Contracting Officer's Representative, and worked on the national long-distance passenger travel forecasting model.
- 2016 to 2021AARP Public Policy Institute, Product Manager and Senior Methods Advisor. Led the AARP Livability Index (50+ data sources across 7 domains, scoring U.S. neighborhoods, across multiple public releases) and co-authored its 2018 report.
- 2024 to nowQNFO (independent research), Founder. An open, AI-assisted research pipeline that runs on 44 deployed Cloudflare Workers (October 2026). Every work carries a DOI, and corrections ship as new versions.
Earlier work is published under Brad Gudzinas. Full CV: doi:10.5281/zenodo.23082080 · ORCID 0009-0002-4317-5604
How I work
AI agents do much of QNFO's engineering, analysis and drafting under my direction. I am accountable for every result, and each deliverable says which parts were AI-assisted.
There is no price list. Scope and fee are agreed for each engagement before any work starts; research collaboration has no fee.
Contact
Email rowan.quni@qnfo.org. The buttons above start the subject with a tag such as [work-with-me:jpcub]. Please keep it: it is how I count which offers bring people here. There is no form on this page; your message arrives in my qnfo.org mailbox like any other email.
New papers by email: subscribe on the home page.
Selected works
- The Joules-per-Solution Metric: Definition, Measurement Protocol, and Anti-Gaming Provisions
- Error Correction Is a Landauer Machine: The Thermodynamic Floor of QEC Overhead
- JPCUB Competitive Landscape v2.0 (17 platforms)
- Joules-per-Solution for Stochastic and Agentic Inference (LLMs)
- The Universal Ignorance Audit
- Epistemic Legibility in AI-Assisted Science
- Operating the Quniverse Fleet: Objectives, Successes, Failures, Roadmap
Quni-Gudzinas, R. B. Published by QNFO. Prepared with an AI-assisted research pipeline; the author is responsible for the content.
This page was prepared with an AI-assisted research pipeline; the author is responsible for the content.