One-time engagement
Technical due diligence
An independent read, in about two weeks, of what an AI company has really built before you invest in or acquire it: the architecture, each AI claim verified or not, the evals, the cost curve at scale, model and key-person dependency.
Know what is really built before the wire.
- About 2 weeks
- Priced per deal
- The report and a partner-level readout
Is this for you?
A term sheet is moving
The demo was impressive, the deck says AI, and someone on the committee has asked how it actually works.
You are acquiring the technology
The price assumes the system is what the data room says it is, and you want that checked by someone who has built one.
You are the founder, and diligence is coming
You want your own read first, under the same discipline, so nothing in the data room surprises you.
Not this? The architecture review is for a company’s own team, on a system it runs; the production readiness audit tests a live agent for the team that owns it.
What you get
- 01
The report
The architecture as built and each AI claim marked verified or not, with the evidence from the data room and the code.
- 02
The risks, ranked
What could go wrong after the wire, ranked by how much it would cost and how hard it is to fix.
- 03
The partner-level readout
The findings in the language of the investment committee, with the questions to put to the founders.
Illustrative pages. Yours carry your system’s scores, numbers and fixes.
The readout
Risks, ranked
- 1CostFix
- 2CostFix
- 3CostFix
- 4CostFix
- 5CostFix
Due diligence report
Six areas, each claim verified or not
- The architecture as built
- Each AI claim
- Eval maturity
- The cost curve at scale
- Model and vendor dependency
- Key-person risk
What I verify
Six things a term sheet quietly assumes about an AI company. I check each one in the data room, in the code and with the team.
01The architecture as built
I read the system as it actually runs, not as the deck draws it, and map what a customer’s request really passes through.
I measureThe architecture as built, next to the one claimed
02Each AI claim
I take every claim the company makes about its AI and mark it verified, partly true or not, with the evidence.
I measureEvery claim, verified or not
03Eval maturity
I check whether the company can measure its own AI: what is tested, on what, and whether the numbers in the deck can be reproduced.
I measureWhat is measured, and whether the numbers reproduce
04The cost curve at scale
I plot what the AI costs to run per customer today and at the scale the plan assumes, so the margin in the model is checked against the bill.
I measureCost per customer at today’s scale and at the plan’s
05Model and vendor dependency
I map what the product depends on, one model, one provider, one API, and what a change in price or terms would do to it.
I measureEvery dependency, and what stands behind it
06Key-person risk
I find who really holds the system in their head, and how much of it is written down, so the price accounts for what leaves with them.
I measureWho holds what, and what is written down
About 2 weeks, the report and a partner-level readout
The two weeks
- Days 1 to 3
The data room
Access under the fund’s terms, the data room and the code read, the claims listed.
- Days 4 to 9
Verify
Each claim checked against the code, the traces and the evals; sessions with the team; the cost curve built.
- Days 10 to 12
Report and readout
The report written, the risks ranked, a partner-level readout, shaped to the deal’s timeline.
The next step
Start with a
20-minute call.
Tell me the deal stage and the date that matters. You leave knowing whether I can take it, how long it would take and what it would cost.
Bring
- The deal stage and the date that matters
- What the company says its AI does
- Any conflict you already see
Why me
I take diligence only where I have no stake in the outcome: no follow-on work with the target, for me or for NeuroGen, the studio I founded, and any conflict named before you brief me. I have sat on the other side of the table too: the surgery discovery platform I built at LetsMD helped the company secure its round. I read a data room the way I read a system I am about to own.
The track record“Mandeep is the level-headed, methodical thinker you want on your team. He is capable of looking at the big picture, anticipating problems and planning before hand.”
Rajat HansCo-founder and CTO, Kenko AIAsked on most first calls
The questions.
How long does it take?
About two weeks from the day the data room opens, shaped to the deal’s timeline. A single product with one model is quicker; a platform with several models and customers takes the full two.
What access do you need?
The data room, read access to the code where the fund can arrange it, a sample of traces or evals if they exist, and the team for a session or two.
Do you sign the fund’s NDA?
Yes. Fund-side work runs under the fund’s own terms.
Where do you stand on conflicts?
I take diligence only where I have no stake in the outcome. No follow-on work with the target for me or for NeuroGen, and any conflict named before you brief me.
Can a founder commission this?
Yes, as preparation before a round, under the same discipline. The report then says what the fund’s diligence will find, so nothing in the data room surprises you.
What if the AI claims hold up?
Then the report says so, claim by claim, with the evidence, which is what the investment committee was asking for.
What does it cost?
It is priced per deal, fixed on the first call before the data room opens.