# Two years as a founding engineer: what we shipped, what didn't work, and my own practice

Maybe you read one of my posts, or you came here from my services page (or we already know each other). Let me introduce myself.

I'm Francisco Luna, a software engineer from Venezuela specializing in full-stack engineering, AI/RAG, and building for pre-seed and seed teams.

From 2024 to 2026, I worked as a founding engineer building from scratch: first, a habit app launched into an existing 1,600+ member community; then, a multi-tenant B2B2C platform built for an 11,500-member gym network :)

* * *

## 2024: Building from zero

I joined as the first engineer, and nothing existed. No infrastructure, no deployment process, no roadmap. I built the foundation, turned the CEO's worries into a roadmap, and we hired our first marketer. I deployed the infrastructure on AWS first, then moved to DigitalOcean because it was cheaper for the stage we were at.

## 2025: Quokka Buddy, and the market

* * *

![](https://cdn.hashnode.com/uploads/covers/689f4f0d610919fe10ba3c12/a8221ef1-bf36-4467-973b-14951085ade9.webp align="center")

Leading a Quokka Buddy product review in November 2025, sharing the live app. I was the first engineer on it and responsible for communicating all the technical changes to the team.

* * *

In 2025, we focused on Quokka Buddy. The idea was simple: you pair with someone and keep each other consistent. Retention never happened inside the app; it happened in a Discord server where people were waiting for each other.

By October 2025, the product was live, and Quokka Buddy placed third in the Business and Tech category of the [IEBS Entrepreneur Contest 2025.](https://www.iebschool.com/hub/ganadores-del-concurso-emprendedores-iebs-2025-digital-business/)

Then we learned an expensive lesson: being good at building is not the same as being needed. Habitica had already solved that problem better, and we kept the entire product free as well. We played it safe, and it didn't protect anyone, including our users.

In November 2025, we hired a second engineer and, for the first time, wrote down how we deployed and who was on call.

That year gave me my philosophy: ask whether the product needs to exist before asking whether I can build it. And put a price on it early enough to hear the answer.

By then, real usage had slowed the app down, so I moved Postgres onto a private network and queries went from 817ms to 119ms, about seven times faster.

* * *

## 2026: FitMate, and the end

In December 2025, two months after launch, we started FitMate. The customer is the gym; the user is its member. The problem is retention, because a gym loses half its members in the first few months.

We used the same tech stack as Quokka Buddy, focusing more on testing this time, with:

Node.js with Fastify, TypeScript, Drizzle, PostgreSQL 16, React 19, Docker, GitHub Actions, Vitest with PGlite, Playwright.

I wrote 141 integration tests on tenant isolation and input validation before the first deploy. In multi-tenant, one tenant seeing another tenant's data is not a bug, but the end of the company.

We onboarded Edge Fitness, a US gym network of 11,500 members, as our pilot brand. V1 reached production there and ran for a short time. It reached people, but it did not convert. Everything else was building and early demos. I learned that a demo is not a launch, and I now define launch criteria before the first commit.

In May 2026, the business changed direction to merge tenants into brands, and in June 2026 the company shut down.

When it shut down, I went back to first principles and focused on the problems I kept hitting myself.

* * *

## June 2026 to now: AI engineering

From June 2026, I went deeper into AI engineering, with Fastify and Node.js. It wasn't because of a trend, but because of a genuine problem I had:

Every week I sent reports and compliance evidence to my CEO, spending two to three hours on information already sitting in GitHub, pasting things into ChatGPT with no way for him to check it himself. So I built Scrapecat. It reads a repository and turns the history into a progress report anyone can read. 18,000 views and 70 upvotes on Reddit.

You can check out the live product here: https://scrapecat-frontend.vercel.app/app

Code: https://github.com/franciscoluna-28/Scrapecat

The second was permissions. Standing up Postgres clusters and VPSes for compliance kept costing me minutes I had not budgeted, so I built Envoy, not the proxy everyone knows: a small tool that tells a team whether its database permissions are right. https://github.com/franciscoluna-28/envoy

I also write for the LatAm and Venezuelan dev community, and one post reached 22k views.

Currently, I am building my own practice, going deeper into AI engineering and automating the unsexy work with agents.

* * *

## Let's work together

I take ownership of what I build, and I won't let AI break production for you. I build SaaS products, AI products, backend systems, and internal tools for new founders and small teams at pre-seed and seed, on contract.

Projects start at $4,000, and I work per project or on a monthly retainer, never hourly.

https://www.itsfranciscoluna.com/ · https://github.com/franciscoluna-28 · https://calendar.app.google/mby4Et5oPoyqBSBF7

If you read this far, thank you. :)