Yves Siqueiraformer CTO of Conpass, acquired by Omie
I show you what the AI decided for you: what needs to change now and what can wait, with your roadmap in mind. I was the CTO of a SaaS from zero to exit, before AI, with 3 million users. And I still love to code.
“The app broke and the AI can't fix it.”
“I don't know what the AI decided for me.”
“The bill keeps going up.”
Vibe coding is the fastest way to validate an idea, and a prototype doesn't need to be perfect. The problem is treating a prototype as a product.
That's why I look at where the business is before looking at the code. A .vercel.app address is fine while you validate. Sending it to customers? Buy the domain. A lot can be improved inside Lovable itself, without throwing anything away.
I tell you what needs to change now and what can wait. And if nothing needs to change, I'll tell you that too.
30 minutes to understand where you are and where it hurts. Free.
I review the system beforehand, match it against your roadmap and, in a 60-minute call, give you a forecast of what to do and when.
Sessions so you can keep going with the AI, or I get my hands on the code with you.
When the product and the team grow: technical direction, rituals and roadmap.
ONE-OFF SESSION
You leave with a written summary of the priorities.
Let's talk →MONTHLY · 3 MONTHS
4 sessions a month and a direct line. For people building with AI who want someone looking alongside them.
Let's talk →HOUR PACKAGE
Me in the code, with the architecture decisions that come with it. From 10 hours a month.
For when you need someone experienced with their hands on the code.
Let's talk →MONTHLY · FEW SPOTS
Technical direction about one day a week: team, rituals, roadmap and hiring. Code included, on request.
For when the product and the team have already grown.
Let's talk →HEALTHCARE · SUPPLY MANAGEMENT
Challenge: around 157 Lambda functions on AWS, many of them duplicated, and three databases scattered around.
Solution: consolidating the functions and the data, with clear responsibilities.
Result: from 157 to 52 functions, and from three databases to one, with a replica.
GOVTECH · PUBLIC TENDERS
Challenge: a bidding bot that acts as the user on public tender portals and has to show everything live to the operator.
Solution: an event-driven architecture: one crawler machine per client and a shared hub feeding the web dashboard.
Result: the operator follows the tender and the bot at work, live, from the browser.
EDUCATION · LMS
Challenge: a spin-off of a company serving large corporations wanted to reach small and mid-sized businesses, with no system and no team.
Solution: a multi-tenant platform built from scratch on the business rules they already knew inside out, plus help hiring the team.
Result: the platform is still live, run by the team I helped build.
On a public tender bot that relied on scrapers over unstable sources, failures only showed up after the damage was done. I instrumented the backend and built a real-time dashboard: the black box became something the team could follow and trust.
A CRM sync started losing contacts intermittently. In the logs, HTTP 429: I was calling the API faster than it allowed. Queue, retry with backoff, idempotency. Today I design it that way from the start, because an unprotected integration doesn't go down: it loses records silently.
Decoupling later costs far more than splitting things right at the start. Hence hexagonal architecture, DDD where the domain calls for it, and multi-tenancy in the database from day one, not in the refactor that never comes.
Spec, then plan, then delivery, with Cursor and Claude in the loop. I always review two things: the tests, because AI covers the happy path and checks what the code already does instead of the business rule; and the architecture, because it delivers something that works and breaks the layers. The speed comes from AI. The judgment is still mine.
Tech advisor & Engineer
Full-stack engineer for over 18 years. I co-founded Conpass, an onboarding SaaS we grew to 3M+ users across 1,000+ companies, until its acquisition by Omie in 2022. As CTO, I wrote the first version of the software on my own and then built the engineering team, which grew from 1 to 10 developers.
I grew up in Florianópolis's startup ecosystem, where brilliant people helped up close. Today I want to be that person for those who are starting out: so that no founder has to make, alone, a technical decision the whole business depends on.
What I learned along the way, and what is now the first thing I set up in any system: observability from day one and well-defined domain boundaries.
Decoupling later costs far more than splitting things right at the start.
Since 2019, I've worked with more than 20 companies as a consultant and tech lead, from healthcare SaaS to public tender platforms. And I still love to code.
APPROACH
Spec-driven, AI-assisted development. Spec → plan → delivery, with Cursor, Claude and modern tooling in the flow. Underneath, hexagonal architecture, DDD where the domain calls for it, and multi-tenancy in the database. The speed comes from AI; the judgment is still mine.
STACK
No. I translate. You leave knowing what to decide, without the jargon.
Not always. A lot can improve inside the tool itself. Migrate only when the moment calls for it.
You tell me about the app and where the business is. I point out what I see and what the next step would be. If working together doesn't make sense, I'll say so.
I can get my hands on the code, with an hour package. But the focus is on you understanding and deciding, not depending on me.
It depends on where your product is and what you need. We work it out in the 30-minute call, with a clear scope and no surprises.
If nothing needs to change right now, I'll tell you.
Let's talk (30 min)The call is free, with no strings attached.