<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Oscar A</title><link>https://oarmas.com/</link><description>Recent content on Oscar A</description><generator>Hugo</generator><language>en</language><lastBuildDate>Tue, 17 Nov 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://oarmas.com/index.xml" rel="self" type="application/rss+xml"/><item><title>If AI Could Read Your Mind, Would You Know What to Ask?</title><link>https://oarmas.com/posts/if-ai-could-read-your-mind/</link><pubDate>Tue, 17 Nov 2026 00:00:00 +0000</pubDate><guid>https://oarmas.com/posts/if-ai-could-read-your-mind/</guid><description>&lt;p&gt;For years, the bottleneck was compute, data, infrastructure. Lately, I’ve been wondering if, for a lot of tasks, it’s becoming us. We keep getting more capable models, agents, skills, plugins and tools that gather more context. And I still find myself struggling to explain what I actually want.&lt;/p&gt;
&lt;p&gt;Try writing down an idea, then explaining it out loud for a few minutes. Sometimes you hear yourself say something you hadn’t even thought to write. You give it more detail, more context, and still look at the result and think: “That’s not quite what I had in mind.” How many times have we done this while vibe coding? You start with a picture in your head and somehow end up building something else.&lt;/p&gt;</description></item><item><title>Building the MLOps Platform</title><link>https://oarmas.com/posts/mlops-from-zero/</link><pubDate>Sat, 20 Jan 2024 00:00:00 +0000</pubDate><guid>https://oarmas.com/posts/mlops-from-zero/</guid><description>&lt;p&gt;Building a platform from scratch is ultimately about &lt;strong&gt;standardization&lt;/strong&gt; and &lt;strong&gt;developer experience&lt;/strong&gt;.&lt;/p&gt;
&lt;h2 id="the-problem"&gt;The Problem&lt;/h2&gt;
&lt;p&gt;How do small companies go from scrappy startup to unicorn-level operations?&lt;/p&gt;
&lt;p&gt;We all know the pattern: early-stage teams, especially in ML and DS, prioritize &lt;em&gt;speed&lt;/em&gt; over &lt;em&gt;robustness&lt;/em&gt;.&lt;br&gt;
Ship features → hit deadlines → experiment fast. Best practices come later.&lt;/p&gt;
&lt;p&gt;But what happens when the company starts raising bigger rounds, scaling operations, hiring dozens of new engineers, and introducing more requirements?&lt;/p&gt;</description></item><item><title>High-Leverage Engineers</title><link>https://oarmas.com/posts/being-top-engineer/</link><pubDate>Wed, 25 Jun 2025 00:00:00 +0000</pubDate><guid>https://oarmas.com/posts/being-top-engineer/</guid><description>&lt;p&gt;Here I will share some of the best advices and things I discovered while I worked at different companies, startups, some unicorns, and some corporates.&lt;/p&gt;
&lt;p&gt;First of all, I am not the kind of engineer who is a LeetCode expert, nor a high-IQ genius. When I was in programming classes, I really struggled with programming tasks, and that’s why I moved to DS. But in the end, I had to learn programming anyway.&lt;/p&gt;</description></item><item><title>The CI/CD Maturity Curve: From Startup Scripts to Enterprise Platforms</title><link>https://oarmas.com/posts/cicd-maturity-curve/</link><pubDate>Wed, 25 Jun 2025 00:00:00 +0000</pubDate><guid>https://oarmas.com/posts/cicd-maturity-curve/</guid><description>&lt;p&gt;Defining strong CI/CD guidelines is an interesting task when building projects or infrastructure from zero.
In this article, I share some recommendations based on my experience working with both startups and large enterprises.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="from-single-repos-to-monorepos-a-practical-journey"&gt;From Single Repos to Monorepos: A Practical Journey&lt;/h2&gt;
&lt;p&gt;When starting a new project, the first decision is usually how to structure the repository:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Single repo:&lt;/strong&gt; All code for one project stored in a single repository.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Monorepo:&lt;/strong&gt; Multiple projects living within one unified repository.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;In early-stage startups, one repo per project is usually the simplest and fastest approach. We keep the code, pipelines, and infrastructure together, and that’s often the right call.&lt;/p&gt;</description></item><item><title>Engineering Culture &amp; Leadership</title><link>https://oarmas.com/posts/first-engineer-latam/</link><pubDate>Thu, 25 Jan 2024 00:00:00 +0000</pubDate><guid>https://oarmas.com/posts/first-engineer-latam/</guid><description>&lt;p&gt;As the first ML Engineer in the region, I quickly adapted to the team and became one of the trusted voices in technical decision-making.&lt;/p&gt;
&lt;h2 id="bridging-the-gap"&gt;Bridging the Gap&lt;/h2&gt;
&lt;p&gt;Joining as the first engineer for LATAM, my initial goal was simple:&lt;br&gt;
&lt;strong&gt;prove the level of software quality and delivery we could achieve from the region&lt;/strong&gt;&amp;hellip; and we did.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="first-task-deploying-a-ray-serving-poc"&gt;First Task: Deploying a Ray Serving PoC&lt;/h2&gt;
&lt;p&gt;My first assignment was to &lt;strong&gt;deploy a PoC using Ray Serve&lt;/strong&gt;.&lt;br&gt;
Not a bad starting point, especially considering that the long-term goal was to standardize all deployments on Kubernetes.&lt;/p&gt;</description></item></channel></rss>