For method & verification
AI you can audit
Someone told you to use AI, and now you're the one who has to defend whatever it produces. The fear underneath that is specific: the model is confident, the output looks finished, and you can't check it. I build models and analyses where every number is traceable and every AI-assisted step is checked against domain knowledge before it becomes a finding. AI drafts and accelerates; it doesn't get the last word. I'm assuming you'd rather have a result you can stand behind than one that arrives fast — if speed is the only constraint, I'm probably not the right fit, and I'll tell you so.
Why now
Anyone can now generate a plausible-looking number, chart, or model in seconds. That's the whole problem. When a convincing answer costs nothing to produce, the convincing part stops being evidence of anything — the scarce thing is a result that still holds when someone with domain knowledge sits down to check it. I won't pretend to know how big that gap is across the market; I'd be making up a statistic to do it. What I can say plainly is that the value has moved from producing an answer to being able to show why the answer is right. That showing is the work I do.
What you get · Fixed scope
Productized offers
A defined scope, a deliverable you own, and a method you can cite. These are the offers built around traceability and verification; the full Services page lists the rest.
Source-traceable causal modeling
quoted quotedTurning a body of literature or evidence into a source-traceable causal model — where each link traces back to a claim, and each claim back to a source, so the reasoning can be inspected rather than trusted on faith.
Independent Replication
4–8 weeks quotedA result rebuilt from the data up by someone with no stake in the original — for when a number matters enough to need an outside check before you rely on it.
Custom dataset + methodology build
quoted quotedYou can't find the data, so I build it — source-traceable, cited, and yours. The dataset comes with its construction method documented, so the numbers are defensible and the process is repeatable without me.
Verifiable-AI workshop
quoted quotedA working session on where AI can and can't be trusted in your pipeline, and how to put verification gates around it — so the checking is a step in the process, not something you hope someone remembered to do.
Why it holds up
Traceable numbers, checked steps
See it demonstrated on a real published figure: One number, fully loaded — a 73% from my ERL paper, unfolded down to its sources and soft spots. Or read a specimen deliverable in the format client work follows.
Verification gates, not vibes
AI output is treated as a draft, never a conclusion. Before anything becomes a finding, it's checked against domain knowledge and against its sources — does this number make physical sense, does the cited claim actually say what the model reported, does the causal direction hold. If a step can't pass that check, it doesn't ship. The gates are part of the method, so verification happens on every result, not only the ones that happened to look suspicious.
Provenance you can walk backward
Every number is traceable to where it came from and how it was derived. The method behind that discipline is published, not a proprietary black box you have to take on trust. That matters when you need to defend a result: you can point to a documented method rather than to my say-so, and someone else can follow the same chain to the same place. Traceable numbers and a published method are what let a finding survive being questioned.
Who's behind it
The work is senior-led — I'm the one on the keys, not a junior team working from a template. My background is a PhD in civil systems engineering and 16 published works, and the method behind the traceability is a published one rather than an internal recipe. Full publication record on ResearchGate ↗.
Bring me the thing you can't check
Tell me what AI or analysis you're being asked to rely on and where the doubt is. I'll tell you honestly whether it needs a replication, a methodology I can document, or just an afternoon putting verification gates around a pipeline you already have — including whether you need me at all.
Other lanes: enterprise & compliance · foundations & mission · researchers & academics.