The hardware was always capable. The software never was.
I've spent 33 years in analytical laboratories — refining, petrochemical, research — and I kept hitting the same wall on every instrument, in every lab. The detector produced rich data, and then the software flattened it, and then we all exported to a spreadsheet to process it the way the chemistry actually demanded. Pivot tables. Copy-paste. A macro from 2014.
Every vendor system is a single-technique silo — Chromeleon, Empower, OpenLAB — built to run one instrument and reluctant to understand anything new. None of them was built for how a chemist thinks. So the richest measurement in the room got thrown away by the software meant to read it. (I go deep on exactly how, for the LUMA detector, in Do You Know LUMA?)
Signal in. Molecule out.
MAI-Alchemy is not another silo. It's a technique-agnostic data platform that turns a raw detector signal into molecular meaning — and it does it in a sequence any chemist can follow:
- Read the signal honestlyStandard, reproducible baseline correction and denoising — the same accepted family every CDS uses — done well, per band.
- Test for coherenceAt every instant, ask whether the 12-band spectral shape looks like a real molecular absorption or like noise. It recovers real peaks height missed and rejects noise — it can never invent a false one.
- Identify by structureMatch the fingerprint across multiple independent axes — spectrum, band ratios, elution order, boiling point — not one score that can lie.
- Compute the propertiesBoiling point, retention index, response factor, density — read from the spectrum, per chemical family.
- Learn from the runEvery analysis sharpens the libraries. The platform gets better the more it's used.
It turns a detector into a virtual lab.
When you can read all of the signal, one injection can do the work of five methods. The same run that reproduces a legacy number — total sulfur, an octane input, a hydrocarbon group — also gives you the full molecular breakdown that legacy method discarded. A short GC column reaches analyses that used to demand a 100-meter column and several hours — once you count the run itself, the peak analysis, and the data processing. That's not a better plot. That's a different capability on hardware you already own.
Statistics, held honest by physics.
A quick word on the tool most modern labs reach for: chemometrics — the branch of math that finds patterns in complex measurement data. Think of it as pattern-recognition for a spectrum or a chromatogram. It's genuinely powerful, and MAI-Alchemy uses it.
But it has one blind spot worth understanding: it only knows correlation. It can find a pattern that fits the numbers beautifully and is still chemically impossible — because the math itself has no idea what chemistry allows.1 That's where a lot of "AI for the lab" quietly goes wrong.
MAI-Alchemy supplies the missing half — physical ground truth — through four things that keep the pattern-finding honest:
Property libraries
Real, measured boiling points, retention indices, response factors and densities for tens of thousands of compounds. When the math proposes an answer, these facts can veto it — a chemically impossible result simply can't survive.
Coherence
A stricter test of whether a signal is real: does its shape hold together across all 12 bands (a real molecule), or fall apart into noise? It reads every channel at once — not just one channel's height.
The σ-wall, named
The point where look-alike molecules give nearly identical spectra — where the spectrum alone would guess wrong. We say so plainly, and break the tie with physics: which compound comes off the column first, and what it boils at.
Run-anchored elution
Every compound is placed by where it appears relative to known anchors in that same run — so the method survives a different column, or a new day, instead of breaking the moment retention times shift.
The shorthand I use: statistics that can't violate physics. The pattern-recognition is the engine; the physical libraries are the guardrails. That pairing is the durable difference — and the part no amount of model-tuning reproduces on its own.
One platform for all analytical data.
It began with the LUMA VUV detector — an innovative technology sitting on information almost no one has leveraged. Unlock the richest, most-overlooked signal first, and the rest of the lab follows. The architecture doesn't care what instrument the signal came from — the roadmap is the whole lab:
And it isn't only for spectral detectors. Even a single-channel trace — FID, TCD — gets sharper peak boundaries and cleaner integration from the same physics-aware baselines and derivative edge-detection, so whatever modeling follows is built on honest data. The point was never one detector. It's reading any signal the way the chemistry demands.
Built by the chemist who needed it.
I'm one analytical chemist with a clear picture of what analytical data processing should look like — and a working, patent-pending platform that proves it. I gravitate toward being first: first to combine five methods on one injection, first to crystallize a compound others couldn't, first to teach the LUMA detector to do what it was always capable of.
What comes next is bigger than one person. If you make instruments, run a lab, or want to help build this future — that's exactly who I'm looking for.
- Chemometrics is correlation-based pattern analysis (PCA, PLS, MCR-ALS); constraining such models with domain and physical knowledge is a recognized need. See R.G. Brereton, Chemometrics: Data Driven Extraction for Science; S. Wold et al. on Partial Least Squares (PLS). Overview.
- Detailed Hydrocarbon Analysis (DHA) and its long-column, multi-step workflow: ASTM D6730. Overview.
- The LUMA VUV detector and its 12-channel spectral output: VUV Analytics. The full technical walk-through is in Do You Know LUMA?
Written from the bench by Medrado Michael Leal. Patent pending.