“Tender Nova exists because I built an engine that keeps AI analytically honed — no drift, no speculation, no hallucination.” — Kevin Cassidy, Founder
“The entire warranty window — May to August 2026 — closed before a February 2027 completion date. It was in the tender the whole time.”
Real finding, real tender, cited to the page. Full documents below — no email required.
“The change-order clause cites Section 3.8.3. Section 3.8.3 does not exist — in any of the three tenders issued from that template.”
“The amendment contained zero language identifying which RFP sections it changed. Bidders had to reverse-engineer it.”
“Two slides, two different corridor limits — and the presentation became a binding addendum with no rule saying which governs.”
“The entire warranty window closed before the job was done.”
“The head schedule’s rows sum to 80. The same table prints 221.”
“A required valve appears on no drawing in the set.”
“The pay mechanism references unit prices. The proposal contains only a lump sum.”
“Flagged the geotechnical data gap before close. Days later, a bidder hit the specific conflict inside it.”
Not one tender run to date has come back clean — ministries, hospitals, universities, municipalities, three provinces.
Findings are statements about documents, checkable against the cited pages; nothing here comments on any evaluation, award, or firm.
Buried in one package: the general contractor, project manager, AND site superintendent each needed references from $2.5M+ municipal renovation projects including abatement — provable within 72 hours of notice, or the bid is non-responsive regardless of price. Liquidated damages of $14,344.79 per working day against a shutdown calendar where extensions had already been refused four times. Late-payment interest struck to 0%. Every cash-allowance request denied.
None of that is hidden. All of it is findable — if someone reads all 300+ pages against each other before you price.
That’s the read: every hard gate, every deadline trap, every risk the contract moves onto you, cited to the page. What you do with it is your call — we identify, you decide.
One rule, always: we work one side of any given tender. If we’ve read it for the issuer, no bidder gets it, and vice versa. First come, first served.
Defects, conflicts, and gaps buried in the documents — a deleted clause still referenced elsewhere, a spec describing the wrong building, two deadlines for the same thing. The kind you find out about after you’ve signed. I read the whole package and hand you every one, cited to the exact clause, page, or sheet, so you can confirm it against the source in seconds. Every finding labeled by how solid it is — nothing wearing more confidence than it earned. You don’t have to take my word for any of it.
And here’s the honest part: how do you know the AI isn’t lying to you? Not tricking — drifting, quietly confident about something it never proved. That’s the whole reason this exists. My method is the discipline that catches the drift — every finding checked against the tender’s own text, so nothing reaches you that the documents don’t back up.
Built and proven on thousands of Canadian tenders · Hamilton, Ontario.
“The AI is the part anyone can rent. The discipline that keeps it honest is the part I built.”— Kevin Cassidy, Founder
Everyone building tender AI is renting the same base models you could rent yourself. The model isn’t the difference — anyone can wrap an AI. Mine doesn’t make the AI smarter; nothing can. What it does is impose discipline: every finding labeled by how solid it is, no guess allowed to pose as a fact, every claim checked against the tender’s own text. It gives you back exactly what’s there — the defects, the conflicts, the gaps — and an honest line around everything it can’t prove from the documents.
Every conflict I find is a mistake someone made writing the tender — so I’m useful on both sides of it. If you issue tenders, I catch the conflicts before they go out: fewer addenda, fewer disputes. If you bid them, I show you exactly what you’re pricing, before anyone else does. I don’t tell you what to bid, and I don’t tell you whether it stands — that’s the engineer’s stamp, not mine. I tell you where the documents don’t agree with themselves — in English, French, and math.
Owner priorities, binding constraints, evaluation mechanics, and unresolved items — organized so the structural reality of the competition is visible before a single bid is opened. No speculation. No probability language. Every claim traces to specific text.
A numbered register of every place the tender text is missing something it should contain, contradicts itself, or creates obligations without defining their boundaries. Five categories: contract risk, evaluation mechanics, document conflicts, scope and pricing, authority and governance. Every finding cited to document and section.
When addenda, Q&A, or clarifications drop during a live tender, the engine runs again — every time, at no extra charge. Each amendment gets its own analysis: what it fixes, what it breaks, what it leaves untouched. Timestamped. Frozen on delivery. How addendum analysis works →
I don’t measure geometry — that’s the engineer’s stamp, not mine. What I read is the language and math written on the sheets: callouts, schedules, notes, revision blocks, figured dimensions. Where a callout contradicts the pay item, where a revision changed the drawing but not what depends on it, where an as-built from another era carries a superseded code — I find the broken logic in what the drawing says, and flag the geometry for the engineer to verify. Every finding cited to sheet and detail. What a drawing logic review covers →
I’m not an estimator or an engineer, and I won’t pretend to be one — I’m a tradesman, seventeen years in heavy civil. Some people understand the new guy on the site better than others.
AI isn’t here to replace the old-guard estimators and engineers. It’s here to validate the old guard — and test their apprentices. The veterans get their work backed by a second set of eyes that never tires. The young ones, who lean on AI without the skillset to check it, get shown exactly where they didn’t verify. My job is the guardrails: putting the discipline and contingencies around raw AI analytics so it can’t drift, guess, or bluff. And it keeps getting better — every time a new model lands, I run it, and the sharper it gets, the more doors my kernel opens. Reading drawings wasn’t possible a year ago. Now, never touching geometry, I catch errors in the prints themselves.
I’ve run thousands of Canadian tenders — heavy civil and engineering — sharpening this model, and not one came back clean. Not one. They run from an embarrassing typo left in while writing, to a wording conflict, to a full contradiction that moves money at bid time — and I tell you exactly which is which. The small ones aren’t filler; they’re the proof. No document is clean — and these days, when an AI wrote it, or an AI read it and passed it, my model still catches what theirs missed. That’s the point: what a human missed, and what another AI missed too.
Then let me prove it.
Send me — or point me to — a closed heavy civil or engineering tender you believe is clean. I’ll show you the breaks in its logic, cited to the clause, ranked from harmless to bid-killing. There’s never been one without a break — and if you wrote it with AI, or an AI read it and passed it clean, all the better: when my model catches what theirs didn’t, that’s the whole thesis proven in front of you. If I somehow show you nothing at all, I owe you a full run at no cost, every amendment followed to close. I’m not selling confidence. I’m betting on it.
Kevin Cassidy · Founder, Tender Nova AI Inc. · kevin@tendernova.ai · 905-541-9118
“Once you can force AI to stay this organized, it stops reading one document at a time. It reads the whole stack against itself — the tender, the SIs, the quotes — and catches where they contradict each other.”— Kevin Cassidy, Founder
An AI is a calculator, and a calculator is only trustworthy because you brought the equation. You don’t ask it what a good number is. You ask it what 47 × 19 is — and you can check the answer, because you already know what a right answer would look like.
A tender package is the same. Hand an AI three hundred pages and no equation and you’ve asked a prediction machine to invent the standard and grade against it at the same time. Of course it makes something up. You gave it no ground to stand on.
What I bring is the equation: a verified structure of what the documents have to contain, what must reconcile, what cannot contradict — and laws the work is held to. Never resolve a conflict, report it. Cite every source. Label anything that can’t be verified. That structure sits outside the AI and doesn’t care how confident the AI sounds.
The how stays under the hood — that’s mine, and it’s what makes this worth something. But you don’t need the blueprint to judge the work. Send me a tender you already know the outcome of, and check what I find against what actually happened. The proof isn’t in me explaining my logic — it’s in you holding the finding and seeing it hold up.
I also proved it on code, where the output either runs or it doesn’t: same AI model, run twice on the same problems — once plain, once with my logic over it. The disciplined version caught real issues the plain pass walked past, and showed its work. Want to see that experiment? Ask me.
“I read what’s written on the sheet — the callouts, the notes, the figures. I don’t measure geometry, and I won’t pretend to. That’s the engineer’s stamp, not mine.”— Kevin Cassidy, Founder
Sprinkler replacement, two occupied towers. Three documents in the same binding package, three different answers to the same question:
Read it one way, you eat a change order. Read it the other, you price a pump and lose on price. The conflict is invisible to anyone reading only one document. I found it in the written content on the drawings, cross-checked against the spec and the pricing form. I didn’t interpret the geometry, and I said so where I couldn’t. Twenty-three documented gaps on that tender; eight serious enough to affect who wins.
Ask, and I’ll send you two complete example analyses — one heavy-civil construction tender, one engineering-services tender with all its amendments — start to finish, every finding traced to its source.
There’s a growing field of AI tools that review construction drawings — some genuinely good at what they do. They read the geometry: where a duct crosses a beam, where a dimension doesn’t match between two views. Real work, real value on a building.
I do something different. What I read is the logic — the language and the math, in the tender and on the drawings too. Where a callout contradicts the pay item, where a change hides inside an addendum that claims it changed nothing, where two binding documents disagree on the same number. And I’m built for a different problem: the tender itself — the procurement package, the bilingual sets, the mandatory gates, the combined-rating evaluations, the amendments that move the goalposts mid-tender. Plan-review tools weren’t built for that.
Some of what those tools do would pair well with what I do — they read the geometry, I read the logic. A builder running both would catch what neither catches alone. Already running another process? Send me the same package and I’ll run my logic-check alongside it, free — you’ll see what each of us catches. Just another honest set of eyes.
“The method is built to be checked; trust is earned. Every finding ties to a line you can go read yourself.”— Kevin Cassidy, Founder
Construction, detail design, standing offers, retainers, inspections, surveys, contract administration — across every major evaluation framework in Canadian public procurement. Dozens are timestamped, frozen before the outcomes were known.
Twin bridge replacements. Flagged a drawing-vs-spec conflict before close. The owner corrected it in a later Q&A.
✓ Validated — owner correctedWater intake civil works. Named a missing schedule and a geotechnical data risk; a bidder raised the exact same schedule three weeks later.
✓ Validated twice — bidder confirmedAll 5 MTO regions, 10 procurement types. Systemic template defects that propagate across regions and versions — patterns no firm bidding one tender at a time can see.
Heavy civil bank protection, tidal river. 61-day window, $3,000/day liquidated damages, archaeological and breeding-bird constraints.
Multi-year pile-drilling standing offer. Explicitly non-binding, no minimum volume over five years.
“Smart is easy to find. Smart that’s been through the ground isn’t — and that’s the part this doesn’t pretend to have.”— Kevin Cassidy, Founder
I don’t tell you what to bid, and I don’t tell you whether it stands. I’m not invested in what your tender is worth — I’m invested in whether it holds together.
Send me any closed or awarded Canadian tender you already know the result of, and I’ll run it in full at no cost — a test you can grade against what actually happened. Issuing instead of bidding? Same offer, on a tender you’ve already put out. One free run per company.
A tenth of the standard rate: full discipline, every finding cited, every amendment tracked to close. If it earns the full rate, you’ll know.
Every finding labeled by how solid it is, every contradiction surfaced, every amendment tracked through closing at no extra charge. An unusually large drawing set may carry a scaled fee — quoted before any work begins, never after.
Send me a closed or awarded tender you already know the result of. I’ll run it free, so you can compare what I find to what your own team caught and what actually happened on award. Don’t have one handy? Just ask — I can pull most Canadian public tenders myself. When you’re ready for a live one, your first is a pilot, every amendment covered, and I’m there to the end. You call, I go to work.
Or email directly: kevin@tendernova.ai