The GIMS gnome
Evan Brown
Systems / Solutions Architect
Results · Evan Brown · Highlands Ranch, Colorado

What changed for the people I built it for.

Four jobs, each with what it cost somebody before, what I did, and what it took afterwards. The same four with the engineering in them, and four more from my own software, are further down for hiring teams.

Four resultsIn plain words, with the number
  1. A testing lab in Michigan
    Equilibrium Scientific

    An inspection two weeks away, three months of work undone, and a pass.

    For a state-regulated testing laboratory whose licence depended on the result

    The lab was working under a licence that depended on passing an inspection two weeks away, with about three months of preparation still undone, so I ran the preparation with their team and they passed. Missing the date would have meant a revoked licence and every result the lab had reported called into question.

  2. Testing labs in Michigan and New Mexico
    Equilibrium Scientific

    The answer in about an hour, not a day.

    For newly founded labs proving each method before they could report

    At the labs I worked with, the chemists spent a full day working up the numbers from each instrument run before anyone knew whether the method passed, so I set it up so the export turns into the finished calculations on its own and the chemist checks them, and the answer started coming back in about an hour. A lab with limited runway found out the same day whether to rerun, change the method or write the report.

  3. Labs reporting in several states
    Equilibrium Scientific

    Results go in once, and the right certificate comes out for the right state.

    For labs whose certificates had to follow a different state’s rules each time

    A lab that reports in more than one state kept a different certificate for each, so I set it up so the results go in once and the right certificate comes out for the right state, and the person at the bench checks it instead of rebuilding it. The compliance work around it ran across seven jurisdictions.

  4. Bio-Techne
    2018–2019

    A half-day check, down to about 30 minutes.

    For a product manager who matched every client’s product list against the catalog by hand

    Every time a client sent over their product list, somebody spent half a day matching it against our catalog by hand, because anything whose wording didn’t match exactly had to be checked one by one. I set it up to run on its own in the spreadsheet they already used, and it came down to about 30 minutes of checking.

If a job like one of these sits in your office, Fix one workflow is the same work at one fixed price, and it starts with the free 20-minute check.

The same four, for hiring teamsWhat was at stake, what I did, the number
  1. Michigan
    ISO/IEC 17025

    Three months behind, passed in two weeks.

    Solutions Architect · Equilibrium Scientific

    A Michigan cannabis testing lab, one of the state-regulated laboratories Equilibrium served, was already reporting results under a licence that was conditional on earning ISO/IEC 17025 accreditation, and it reached the last two weeks before the assessment with about three months of preparation still undone. Missing the date would have meant a revoked licence and every result it had reported called into question, with potential lawsuits behind that, so the work could not slip and could not be skipped.

    I took the remaining preparation and compressed it into those two weeks with the lab’s team, lining up what the bench actually did with the procedures, the records and the staff readiness an assessor checks, until the written process and the real one were the same. The lab passed and kept its licence, and across every engagement since, no client of mine has lost an accreditation.

  2. LC-MS · GC-MS · ICP-MS
    Validation

    Validation analyses in about an hour, not a day or more.

    Solutions Architect · Equilibrium Scientific

    A new lab has to prove each method works before it can report results, and the proof runs on data from LC-MS, GC-MS and ICP-MS instruments. Nobody had done that systematically at the labs we served, which were being founded as the work began, and working up one run by hand took a chemist a full day or more, so a lab burning through its runway waited a day to learn whether to retest, revise the method or write the validation report.

    I designed the validation architecture that takes the instrument export and does the processing, so the chemist checks the result instead of building it, and we used it in every engagement from then on, at labs in Michigan and New Mexico among others. The same analyses came back in about an hour. Alongside it, the validation packages I published were accepted by state auditors in more than ten testing categories, including microbiology, potency, residual solvents, heavy metals and mycotoxins.

  3. Certificates of analysis
    Seven jurisdictions

    One set of results, the right certificate in every state.

    Solutions Architect · Equilibrium Scientific

    A certificate of analysis has to follow the reporting rules of the state it is issued in, so a lab working across jurisdictions ends up with a different certificate for each one and an operator who has to remember which is which. I designed and built the certificate generation system in TypeScript, with each jurisdiction’s requirements held in a schema-driven template, so the operator enters the results once and the right certificate comes out for the right state with nothing extra to remember. The compliance work it belongs to ran across seven jurisdictions: Colorado, New Mexico, South Dakota, New York, Akwesasne, Michigan and the Caribbean.

  4. Bio-Techne
    2018–2019

    A half-day catalog check, down to about 30 minutes.

    Data Analyst · Bio-Techne

    Whenever an OEM client sent over the list of every antibody they offered, a product manager had to find which ones Bio-Techne already carried. The check ran on VLOOKUP, and any product whose wording didn’t match exactly had to be checked by hand, so each list cost at least half a day, as often as lists came in. I wrote an Excel macro that checks everything against everything, then optimised it until a whole list ran in about 30 minutes, and standardised the cleanup steps around it. What was left for the product manager was the checking, not the matching.

Across Equilibrium’s engagements I was the technical authority for validation, reporting and accreditation at laboratories in seven jurisdictions, took three labs through state and ISO/IEC 17025 accreditation (two in Michigan, one in New Mexico), published validation packages accepted by state auditors in more than ten testing categories, and generated $600K+ in consulting revenue through the validations, training and engagements I led, with no client accreditation lost.

Engineering proofHow I work when nobody is checking

None of these has a paying customer outside my partner’s business yet, so they are proof of method rather than business cases. Each one links to its evidence.

  1. GIMS
    2025–2026

    A 100K-line application refactored in four days without changing its behaviour.

    GIMS had grown past 100K lines with duplicated backend logic, and its controls for audit readiness needed strengthening. I refactored it on one branch, refactor/foundation: 152 commits across 533 files in four days, each step proven behaviour-neutral before it was committed. What made that safe is that routes are defined by decorators and the guard baselines are keyed by path and name rather than filename, so moving code and rewriting its imports is provably wiring-neutral. The suite stood at 3,544 passing tests on main in October 2026.

  2. autoSQL
    2026

    Two engines that must agree, or no number shows.

    autoSQL compiles a user’s dashboard picks into SQL that runs inside Postgres, then runs the same picks through the original Python evaluator and compares every row, and if the two disagree the screen says so instead of showing a number. Over 11,367 tested expressions the two have 0 disagreements, with the contract fixture at 130 of 130. The first performance spike measured the compiled path at 3.79 to 7.15 times slower than the Python path, and 18 of 33 ways the engines could disagree were undetectable at query time, so I ruled no-go on funding that architecture on that evidence and wrote down the two experiments that would change the ruling.

  3. The factory
    September 5, 2026

    This site, rebuilt in under three hours through two gates I cleared.

    My development runs as a ticket factory: intake, spec, build, an automatic review by a fresh agent, verification, and two gates only I clear. The ticket to rebuild this site went in at 04:35, I cleared the spec gate at 04:59, the reviewer sent the build back once at 06:31 with its reason, and the merge gate cleared at 07:23, every step on the ledger. GUTS, the coordination layer under it, has been the shared foundation for roughly a dozen tools and applications, with the process versioned next to the code and the model at each stage a setting.

  4. GEDS
    2026

    The plot-twist detector a ruler beat, and I published it.

    GEDS watches a story’s scenes and flags the one that changes the structure underneath. With a hand-written brief per story it caught 40 of 40 planted twists and ignored 20 of 20 decoys. Without the brief, over 40 stories, it found 64 of 91 twists and fired on 30 of 40 decoys, while a rule that flags the two longest scenes found 63 and fired on 11, because the AI that wrote the test stories gave planted scenes more words. Thirty-odd rounds were pre-registered, so the loss was measured rather than explained away, and the full write-up is the HackerNoon article “I Built an AI Plot Twist Detector. A Ruler Beat It.”

A workflow to fix, or a team to join

Which of these looks like your week?

Walk me through the job in a free 20-minute check, and you’ll hear what I’d build, how long it would take and what it would cost.