I didn't come to software the short way.
For about a decade I worked as an architect in Buenos Aires — first at an engineering firm, then at a studio, then on my own, running small residential projects end-to-end: clients, budgets, construction crews, the whole thing. I learned how to take an idea from a rough sketch to something real that someone was going to live in. That part stuck with me.
In 2021 I decided to change careers. I was 34, self-employed, and the work I was doing had stopped making sense to me. Switching tracks that late is the kind of decision you can only justify in retrospect. At the time it mostly felt like jumping.
The first year and a half was rough in a way I won't pretend it wasn't.
I studied on free platforms while still taking architecture work to pay bills. I did exercises I didn't understand. I'd write a few lines of code and get stuck for hours. I enrolled in a data science bootcamp and got halfway through before pausing it. I could feel that I was grasping the ideas, but the distance between understanding something and being able to put it into working code was enormous, and most days it felt like I was barely moving.
Then the tools changed.
Copilot arrived, ChatGPT landed, and the loop between "I have an idea" and "I can test it" got dramatically shorter. For someone new to the field, this was a different experience than it was for experienced engineers. I had no habits to unlearn. I could build my workflow around these tools from day one, instead of bolting them onto an older way of working.
What Copilot specifically revealed to me was that I understood more than I could produce on my own. I'd been confusing two different gaps — comprehension and output — and treating them as the same problem. Once I could iterate faster, I started seeing my own reasoning more clearly. I could try an idea, see it run, adjust, and move on, instead of losing hours fighting syntax while the actual thinking evaporated.
My learning curve changed shape. It stopped feeling like climbing a wall and started feeling like running experiments. And experiments were something I already knew how to do — I'd been doing them on construction sites for years, just with different materials.
The project that opened the door
Around that time I tried to build something real: a document extraction tool for a friend's business. I'd attempted it before and hit a wall I couldn't climb. This time, with the new tools, I could actually make it work. I spent months iterating — scoping the architecture, wiring up GCP Document AI processors, writing async Python to coordinate the pipeline, handling edge cases in messy PDFs. It never reached production, but it ran.
The thing I didn't realize at the time was that this half-finished project was going to be the thing that got me my first job. The architecture sketches I'd drawn for it — naive as they were — showed enough systems thinking that a small startup called Innuvio took a chance on me.
What I built at Innuvio
At Innuvio I was one of two developers working alongside our CTO on EchoVox, an enterprise conversational AI platform handling around 3,000 conversations a day across voice, WhatsApp, and web. Within three months of joining I was shipping features end-to-end — real code, real users, real production traffic.
In roughly a year and a half I owned features on things I'd never touched before: OAuth 2.0 flows with Meta, event-driven report pipelines on AWS, WhatsApp Business API integrations, LLM fallback systems. I wasn't handed a curriculum. I had a CTO, a Jira board, and production traffic. I learned by shipping — and the fast feedback loop that had carried me through the self-teaching phase turned out to be exactly the loop I needed at work, just with higher stakes.
Earlier this year the founders decided to wind down the project after a strategic reassessment. Two weeks, and it was over. That's how early-stage startups work. I'm now looking for what's next.
Why this blog exists
This is where I'm going to think out loud about the things I've built and the things I'm building. Not a résumé in long form — a working notebook. Each post will be about one concrete decision: why I chose this architecture, what broke, what I'd do differently, what the tradeoffs actually felt like from the inside. Some posts will be deep dives into EchoVox features. Others will be new projects I'm using to learn things I didn't get to touch at Innuvio.
My background is unusual and my slope is steeper than my starting point. That's the bet I'm making, and this is where I'm showing the work.