Lucas, in his own words
Two minutes, recorded for this search. Told afterwards that it ran long and that he had described two things inaccurately, he re-recorded the same night inside the limit and corrected both himself on camera.
Where Lucas is strong
The dimensions we score every candidate on. The dashed outline is what we judged this particular role to need. Everything further down the page is the evidence behind these positions.
AI engineering was not a requirement for this job and he has it anyway, which is the spike outside the line. Experience and communication are the two places he sits below it, and both are covered further down.
What the exercise actually tested
Every candidate got the same challenge, built for this client and this role and never used before. A real trip with ten priced lines, the founder’s own mockup of the screen, and a brief with a deliberate contradiction in it. Build the pricing screen a travel consultant would actually use. One loss was hidden inside a healthy-looking total, and the answer key was written before any submission was opened.
The loss was hidden inside a healthy-looking total
A flight in the trip was discounted below what it cost to buy. The trip total still reads a healthy 12.4%, because the loss is buried inside it. The transport section is really running at 1.2%. A screen that hides that is a screen that loses money quietly, and this was the whole point of the exercise.
- The section at 1.2% marginHis screen surfaces it. He was the only one of the three whose screen showed it at all.
- The flight sold below costMarked on four separate layers of the interface, so a consultant cannot scroll past it.
- A quote that breaches the client’s budget ceilingWarns loudly and still lets the consultant proceed. The other two either stopped the workflow dead or changed the price without saying so.
- A contradiction inside the founder’s own mockupHe followed the brief, built a drawer explaining the distinction, and wrote a note so the next engineer does not quietly undo it.
Every figure on his screen was checked by hand against our ground truth. All ten lines, all three section subtotals and the trip total match to the cent, including one figure where our own brief carries a typo and he gets it right anyway.
9 The technical challenge9/10 · the strongest of the three submissions Read moreSee less
The short version. His money is right everywhere, his screen shows the thing the exercise was built to hide, and he is the only one who shipped tests, continuous integration and written decision records without being asked.
What it got right
- Every figure exact. All ten lines, all three section subtotals and the trip total match our ground truth to the cent, including one figure where our own brief has a typo and he gets it right instead.
- The screen does the job it was built to test. He is the only one of the three whose interface reveals that the transport section runs at 1.2% while the trip total reads a healthy 12.4%.
- Live recalculation, verified by hand. We edited four states from a 15% discount to a 200% markup and watched it cascade correctly through line, section, trip and headroom every time. The strongest demonstration of the three.
- He handled the case the brief deliberately left unanswerable. When a quote goes over the client’s budget his screen warns loudly and still lets the consultant continue. The other two either stopped the workflow dead or changed the price without saying so.
- He knew where to stop. Asked about a change that would touch the data model, he designed it on paper and did not build it, and said exactly why. Decide on screens, escalate on the data model.
- The repository is the separator. Unit tests, browser tests, continuous integration on every change, and written decision records. None of it was asked for and neither of the other two produced any of it. The pricing module is pure, with every rule documented in its header.
- He volunteered his own uncertainty on camera about a design decision he had made, leaving the founder room to overrule him.
- He made his AI usage inspectable, committing with his assistant named as co-author, so it is visible in the history.
What to weigh
- He set a business rule nobody specified. The 15% blended margin floor is his number. His system warns and leaves the decision with the consultant, which is the right execution, and his own architecture document says thresholds should be configurable values, so if it is hard-coded the build contradicts the doc. Small, and worth one question.
- He cut mobile responsiveness and declared it honestly in his note. Another candidate shipped a mobile layout in the same two and a half hours.
Background and fit
Technically strong is the easy half. We check the same six things on every candidate, because the reason placements fail is usually one of these six, and only rarely raw ability.
Stack
Next.js, NestJS, PostgreSQL and Prisma, Redis and BullMQ, Socket.IO. He has shipped multi-tenant architectures with row-level security and distributed background workers, which is the shape most early products grow into.
Industry
No travel. His domain is B2B platforms and industrial software. Different words, and the same problem of getting money right on a screen somebody has to trust.
Early stage
Comfortable without a specification. From a pricing screen alone he wrote a two page architecture document describing, unprompted, roughly the system the client was actually building.
Leadership
A senior individual contributor, and not a lead. What he does instead is write for whoever comes next. One of his decision records exists purely to stop a future engineer correcting a number that is deliberately right.
Working on existing code
He refactored a legacy editor into a modular architecture with more than 85 automated tests, delivering a platform version in ten days across thirteen pull requests and three repositories. The instinct to rewrite from scratch is absent, which is what a small existing codebase needs.
Communication
Watch his videos before you decide. His written reasoning is the sharpest of the three. He speaks fluent English with a Brazilian accent, and some people tune into it faster than others. Both of his videos are on this page, so you can judge it in about a minute.
8 Background and fit in full8/10 · excellent engineering judgement, with less depth than his work suggests Read moreSee less
The short version. His judgement is ahead of his years, and the years are the honest limit. Nearly six of them, which is less than his work makes you assume.
What he has built
Multi-tenant architectures with row-level security, distributed background workers on Redis and BullMQ with real-time progress in the front end, and autoscaling container infrastructure. He also built a cost-tracking and billing reconciliation system for generative AI that holds 0.4% variance between recorded and billed cost, which is the same discipline a costing engine needs, measured against ground truth.
He writes things down
The habit that shows up everywhere in his work is documentation aimed at the next person. Decision records explaining why a number is what it is. A note flagging which choices were deliberate so a reviewer reads them correctly. A header on the pricing module listing every rule it obeys. On a two person team with a non-technical founder, that is worth more than it looks.
He is dual-qualified, and we tested it
He also sat the AI engineering challenge in this search and scored in the strong band on both, which nobody else managed. Every figure on the AI exercise was correct and his presentation of the results was the clearest of anyone, written for someone who does not read code. The honest limit there is that the hardest step, deciding which supplier price applies to which booked item, he did by hand and did not automate, and he said so in writing before anyone asked.
Where the ceiling is
He scores 8 here and not higher for two reasons. Nearly six years is less experience than his output suggests, and production incident ownership and scope negotiation under pressure are both untested. The second is the spoken communication risk in the grid above. Neither is a reason against him. Both are reasons to run the call properly.
What we checked
- His CV and his LinkedIn were read against each other and against the job description. The specific claims are granular and mutually corroborating. His public profile rounds his experience up slightly, well inside normal CV rounding, and we are stating the lower figure here.
- His challenge repository was cloned and read directly, including the pricing module, the decision records and the architecture document.
- Every figure on his screen was checked by hand against our ground truth, and the interface was edited through four states to confirm the recalculation.
- Both of his walkthrough videos were watched in full.
The short version
What stood out
- His screen showed a loss the other two hid. One section of the trip was running at 1.2% margin while the total still looked healthy at 12.4%. Only his interface surfaced it.
- Every figure was correct, to the cent. Including one where our own brief carries a typo and he gets it right anyway.
- He was the only one who shipped tested code. Automated tests, browser tests and checks that run on every change. None of it was asked for and neither of the other two produced any.
- He also passed our AI engineering challenge. Nobody else in the search passed both.
- He knows which decisions are not his to make. A change that would have altered how the product stores its data he designed on paper and left for the founder to approve, and said why.
- He makes his AI use visible. He committed with his AI assistant named as co-author, so how the work was produced is there in the history.
What to weigh
- Watch his videos and check you follow him easily. His English is fluent and his written work is excellent, and accents suit some ears better than others. Both of his videos are on this page, so it takes about a minute to judge.
- Nearly six years is less experience than his work suggests. Owning a production incident and negotiating scope under pressure are both untested.
- He made one commercial decision that was not his to make. He picked the profit threshold the screen warns at, and nobody had given him one. He warns and does not block, which is the right instinct, and numbers like that are worth agreeing up front.
- He prioritises under time pressure and tells you what he dropped. He left the screen desktop-only and said so plainly. Another candidate fitted a mobile layout into the same window.
- No travel industry experience. He would be learning the vocabulary on the job.
Hire Lucas, or someone vetted like this
Shortlists are free. Tell us the role, we run this process on it at our own cost, and you get the write-ups whether you hire anyone or not. If you hire Lucas off this page there is no fee, to you or to him.