Four for the lead AI role
Emmanuel has done the full technical exercise and is ready for you to meet. The other three have been interviewed but not tested yet, and I would rather not spend a few hours of their time before you have looked.
Prepared 2 September 2026. This sits alongside the first shortlist rather than replacing it.
Do you want to meet Emmanuel?
He has done the exercise and been graded against the same answer key as the last round. He is the one I would put in front of you.
Should I send the exercise to anyone else?
Kevin, Igor and Emelian have all been interviewed. Tell me which of them, if any, and I will send it. One line is enough.
Ready to interview
Emmanuel Abugauch
Córdoba, Argentina · 8 years · Head of AI at a payments company operating in 80 countries
He has built this system before, commercially. At Yuno, supplier terms arrive by website, PDF, email and phone call, and his team turns them into one clean catalogue that pricing runs against.
Available Part-time alongside his current role · start date to confirm
9score$50/hr$4,333/mo at 20 hrs/week$2,167/mo at 10 hrs/weekRecommendedSee moreSee less
Emmanuel Abugauch
Córdoba, Argentina · 8 years · Head of AI at a payments company operating in 80 countries
He has built this system before, commercially. At Yuno, supplier terms arrive by website, PDF, email and phone call, and his team turns them into one clean catalogue that pricing runs against.
Available Part-time alongside his current role · start date to confirm
What stood out
- He has done this exact job before. Same problem shape, different industry, at a company operating across 80 countries.
- The strongest exercise we have graded. It found everything hidden in the documents, and it kept behaving correctly when I changed the inputs without warning.
- He reports his own system’s faults rather than hiding them. That is the habit that keeps a wrong price from reaching a client.
- He runs a twelve-person AI team and still writes the hard parts himself, so he can both make the calls and build.
- Comfortable in work where being wrong costs money. He built the system that protects card data at Yuno.
What to weigh
- He is easier to follow in writing than in conversation. He pauses and restarts, and both videos open with how the system works rather than what it produced. The substance is there, it arrives late.
- He thinks this kind of system should eventually run without a person signing off, earning trust through measurement instead. Considered, but the opposite of approve-before-sending. Worth hearing him out on it.
- In his video he says “my tests are wrong.” He means he ran out of time to close gaps his own tests exposed. The suite passes, I ran it. Read that line in context.
- He is Head of AI at a funded company, so this would sit alongside that.
9The technical exercise9/10 · the strongest of the submissions we gradedSee moreSee less
The short version: he found every problem hidden in the documents, his figures reconcile exactly against our answer key, and he reported a fault in his own work that nobody asked about.
What he found
Seven problems were deliberately hidden in those documents. He found all seven, including the one that matters most: a villa the client booked that has no price anywhere. His quote comes to $30,140 and he says plainly that it cannot be sent. Five services have no price at all, and one of them stops the whole thing.
The test that separated him
I changed his source data four ways without telling him: renamed a room type, deleted a service, added one that did not exist, and moved the whole trip forward a year. Each time it did the right thing. Nothing was quietly dropped, nothing was quietly charged, and when the dates moved past the end of the rate sheet it flagged every line rather than reusing last year’s prices.
The moment worth knowing about
One line in the rate sheet is damaged, so the text comes out garbled. His system worked out the right price anyway, then wrote a tidy quotation of a source that does not exist in the document. The AI got the number right and invented the evidence for it.
He found that himself, built a check that detects it, and opened his write-up with it. He is the only person in this search who has reported a fault in their own work.
One line in his video to read in context
He says “actually, my tests are wrong.” What he means is that he wrote his tests first, they exposed gaps, and he ran out of time to close all of them. The suite passes and I ran it. He is pointing at his own published accuracy report, which shows where his system scores badly. It is a disclosure rather than a defect, but it sounds worse than it is.
How he handles being unsure
Every price is sorted into one of five states: confirmed, replaced by a later email, assumed, based on an expired rate, or not found. The label is not a judgement call. It comes from six yes-or-no facts anyone can check. Every figure names the document and the row it came from, so you can verify a number without reading any code.
9Background and fit for this role9/10 · he has built this system before, commerciallySee moreSee less
What he has already built
He leads AI at Yuno, a payments company operating across 80 countries. Payment providers each publish their terms differently. One has a decent website, one sends a PDF, one replies by email. His team built something that reads those sources, turns them into one structured catalogue, and tests each entry against the provider’s real system until it works. Adding a new provider went from about five weeks to hours.
Why that matters here
It is your problem in a different industry, and the important half is the second one: pricing never touches the original documents. It runs against the checked catalogue, offline, with no AI involved. That is the design that stops a made-up number reaching a client.
He also built the system that protects card data at Yuno, which is what allowed the company to be certified to handle payments at all. He is used to work where being wrong has consequences.
The thing he says in his video that stuck with me
“It is really hard to create a prototype and move this to production. In test you can see a PDF with a few tables in one format, and in production you get another PDF with other types of columns and other tables.” That is precisely where you are: something works, and then real supplier documents arrive.
How he comes across
His English is fluent and he is easy enough to follow, but he pauses and restarts more in conversation than in writing, and he tends to explain how something is built before saying what it produced. Both videos open with architecture rather than the result. The substance is all there, it just arrives a minute or two later than you would want.
The one thing to put to him directly
He argues a system like this should eventually run without a person signing off each time, earning that trust through measurement instead. It is a considered position with a real argument behind it, but it is the opposite of approve-before-sending, and it is worth hearing him defend it before you decide.
Emmanuel against what this role needs
The dotted outline is what we think the role needs. These positions are our judgement, offered as a summary of the assessment above.
Interviewed, not yet tested
All three are strong enough that I would happily put them in front of you. What they do not have yet is a few hours of real work graded against a known answer, which is the thing that separates people properly.
Kevin Wolf
Costa Rica · 16 years · acting CTO of a live US product · full overlap with New York hours
He runs a live product where an AI answers the phone, takes a stranger’s details, produces the paperwork and collects payment, with a person approving before any money moves.
Available Immediately
8score$70/hr$6,067/mo at 20 hrs/week$3,033/mo at 10 hrs/weekFastest to moveSee moreSee less
Kevin Wolf
Costa Rica · 16 years · acting CTO of a live US product · full overlap with New York hours
He runs a live product where an AI answers the phone, takes a stranger’s details, produces the paperwork and collects payment, with a person approving before any money moves.
Available Immediately
What stood out
- He is easy to follow. Clear, structured, no wandering. Worth watching his video first given what you said about the last round.
- Sixteen thousand agencies, every one with its own paperwork. His system reads documents it did not design and works out where each piece of information belongs. That is the closest thing in his record to your supplier rate sheets.
- The AI scores and stops; a person approves before money moves. He can explain why the line sits exactly there, which is the reasoning you want behind that decision.
- He owns a live product end to end as the sole engineer, and delivered it six weeks ahead of his contract.
- Sixteen years, and the closest fit with your existing code of the four here. He hosts on Vercel by preference but says he will work with whatever the team picks.
What to weigh
- How he checks his own accuracy is the open question. His systems have clear guardrails, but nothing in his record shows a method for measuring whether the output is right. That is exactly what the exercise puts under a clock.
- His AI work mostly produces things, like speech and documents, rather than pulling figures out of a supplier’s pricing. The document work he does have is about placing information into forms rather than extracting money out of them.
- The most expensive of the four, by $10 an hour.
8Background and fit for this role8/10 · clear communicator, strong ownership, untested on accuracySee moreSee less
What he has built
He is the engineer behind a product used by bail bond agencies in Utah, working as its acting CTO. Someone calls, an AI answers, takes the details, looks up the case, scores the application and produces the paperwork. A person approves it, and only then is a payment link sent.
The part that maps onto your problem
There are around sixteen thousand of these agencies in the US and each has its own forms. His system reads a document it has never seen, works out what each field is for, and places the case information into it. That is the same shape as reading a supplier’s rate sheet you did not design, though it stops short of pulling a price out of one.
The answer that stood out
Asked where a human sits in the chain, he said the system scores and stops. It does not act, because acting on some of those grounds would be illegal. He arrived at approve-before-money-moves through a real constraint rather than a design preference, which is why I believe he would hold that line here.
What the exercise would settle
Two things. Whether he can pull structured figures out of a supplier document under time pressure, and whether he has a way of checking his own accuracy. He is available now, so he could start on it as soon as you say the word.
Kevin against what this role needs
The dotted outline is what we think the role needs. These positions are our judgement, offered as a summary of the assessment above.
Igor Pimenta
Belo Horizonte, Brazil · 10 years · founder of his own AI company · former product manager
Runs his own AI company and was a product manager before that, so he is used to owning the decision as well as the code.
Available About one week’s notice
8score$50/hr$4,333/mo at 20 hrs/week$2,167/mo at 10 hrs/weekClosest to a founderSee moreSee less
Igor Pimenta
Belo Horizonte, Brazil · 10 years · founder of his own AI company · former product manager
Runs his own AI company and was a product manager before that, so he is used to owning the decision as well as the code.
Available About one week’s notice
What stood out
- He built a system pulling car pricing, specifications and weekly promotions out of PDFs, spreadsheets, web pages and emails into one place, for Brazil’s largest dealership group.
- Those promotions had eligibility rules checked against each customer. Offer one wrongly and the deal was lost. That is your margin problem in another industry.
- He engages with your product rather than his own CV. In his video he calls your current workflow stable but manual and impossible to scale, then asks you what your biggest concern is.
- Asked whether he had kept separate companies’ data apart, he said he might not have, then described exactly the right approach. I would rather that than a confident yes.
- His reason for wanting this is specific: agency projects end, and he misses seeing a product grow past launch.
What to weigh
- No formal way of measuring whether the AI is right. He watches for problems rather than testing against known answers, which is the difference between noticing something looks wrong and knowing that it is.
- He explains at length rather than leading with the answer. Given what you said about the last round, this is the one to listen for.
- Would need about a week before starting.
8Background and fit for this role8/10 · real founder judgement, no formal measurementSee moreSee less
The dealership system
Four kinds of source, updating constantly, with one brand changing its prices twice a day. Emails arriving in a fixed format were captured automatically; documents came from a shared drive the dealership staff maintained. Everything was indexed so a sales assistant could answer a customer question in seconds.
The part that maps onto your product
Promotions varied by dealership, by location and by date, and had to be checked against each customer before being offered. Get it wrong and the sale was gone. That is the same shape as applying the right rate to the right night for the right supplier.
Keeping clients apart
At an agribusiness client he built the permissions so that one customer could never reach another’s documents, enforced before any query runs rather than in the application. He raised that himself, and he raised it in terms of what a determined user could get at rather than what the screen shows.
What he is missing
He built a second AI to watch the first one and report problems to a dashboard. That is monitoring. What he does not have is a set of known-correct answers to test against, which is what the exercise would give us.
Igor against what this role needs
The dotted outline is what we think the role needs. These positions are our judgement, offered as a summary of the assessment above.
Emelian Nichiforov
Mexico · 7 years · staff AI engineer at a top US cancer centre
Builds document-processing AI where researchers act on what the system tells them, so being wrong has a cost. The deepest document work of the four.
Available Four hours a day, weekends if needed
8score$60/hr$5,200/mo at 20 hrs/week$2,600/mo at 10 hrs/weekStrongest on accuracySee moreSee less
Emelian Nichiforov
Mexico · 7 years · staff AI engineer at a top US cancer centre
Builds document-processing AI where researchers act on what the system tells them, so being wrong has a cost. The deepest document work of the four.
Available Four hours a day, weekends if needed
What stood out
- A system reading over 100,000 scientific papers so researchers can find and verify evidence in minutes rather than weeks.
- His way of checking accuracy is the most rigorous here. His test cases come from weekly calls with the scientists using the system. They tell him what the right answer was, and each one becomes a permanent test.
- He found a flaw by deliberately feeding his system a document larger than anything it had seen, watching it break, and fixing it.
- He asked, unprompted, whether the exercise needed a human review step, before I had said anything about how your product works.
What to weigh
- He overstated one area of his experience. Asked how he had kept one client’s data from reaching another, he said he had two years of it, then described two systems that were not that. Everything else he told me checked out, but it is the one answer I would want tested rather than taken on trust.
- He works in Python day to day rather than TypeScript, which is what your front end uses. The AI layer can sit in Python and your product engineer owns the interface, so this is workable, but he would not be the one reviewing front-end code.
- Seven years inside a consultancy on client placements rather than building something of his own.
8Background and fit for this role8/10 · deepest document work, one answer I would testSee moreSee less
Why his accuracy method matters
Most engineers check their AI by comparing it against what their own system produced earlier, which only proves nothing has changed. His comparison points come from the scientists who know the right answer. That is the difference between a system that is consistent and one that is correct.
The reservation, plainly
I told him in advance I would ask how he had kept one client’s data from reaching another. He claimed two years of it, then gave two examples that were something else. He is strong on measurement and I would test that particular claim rather than take it on trust.
How he would find the time
Four hours a day, weekends if there is a deadline, alongside a current role he describes as quiet at the moment.
Emelian against what this role needs
The dotted outline is what we think the role needs. These positions are our judgement, offered as a summary of the assessment above.
Side by side
The chart shows Emmanuel against what we think the role needs. Tick any of the others to lay them over the top, including Barbara and Andrei from the first shortlist.
Emmanuel against what this role needs
The dotted outline is what we think the role needs. These positions are our judgement, offered as a summary of the assessments above.
| Emmanuel | Kevin | Igor | Emelian | |
|---|---|---|---|---|
| Has built this shape of system before | Yes, commercially | Partly | Yes | Yes |
| Pulls figures out of documents he did not design | Yes | Reads them, does not price them | Yes | Yes |
| Has a way of checking the AI is correct | Yes, measured | Untested | Monitoring only | Yes, measured |
| Keeps separate clients’ data apart | Yes | Yes | Yes | Claimed, unverified |
| Has owned a product end to end | Yes | Yes | Yes | Client placements |
| Comfortable in your front-end stack | Reads it | Yes | Prefers Python | No |
| Technical exercise | Done, 9/10 | Not sent | Not sent | Not sent |
| Could start | To confirm | Immediately | One week | Immediately |
What each would cost
All-in. There is nothing else to pay on top.
| Engineer | Hourly | 20 hrs/week | 10 hrs/week |
|---|---|---|---|
| Emmanuel Abugauch | $50 | $4,333 | $2,167 |
| Igor Pimenta | $50 | $4,333 | $2,167 |
| Emelian Nichiforov | $60 | $5,200 | $2,600 |
| Kevin Wolf | $70 | $6,067 | $3,033 |
Two months at 20 hours a week and then 10, which is the plan we discussed.
What I would do
Meet Emmanuel. He produced the best piece of work in this search, at the rate you originally wanted, and he has built this same system commercially at a payments company. Watch his two videos first: he is easier to follow in writing than in conversation, and you should judge that for yourself rather than take my word for it.
- Book Emmanuel this week. Half an hour is enough to know.
- Tell me who else should get the exercise. Any of the three, or none.