<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" ><generator uri="https://jekyllrb.com/" version="3.10.0">Jekyll</generator><link href="https://maxrios.dev/feed.xml" rel="self" type="application/atom+xml" /><link href="https://maxrios.dev/" rel="alternate" type="text/html" /><updated>2026-06-10T04:11:33+00:00</updated><id>https://maxrios.dev/feed.xml</id><title type="html">Maximiliano Rios</title><author><name>maxrios</name></author><entry><title type="html">Record Listening Club - Today’s Listens</title><link href="https://maxrios.dev/2026/02/07/Record-Listening-Club-Albums/" rel="alternate" type="text/html" title="Record Listening Club - Today’s Listens" /><published>2026-02-07T00:00:00+00:00</published><updated>2026-02-07T00:00:00+00:00</updated><id>https://maxrios.dev/2026/02/07/Record-Listening-Club-Albums</id><content type="html" xml:base="https://maxrios.dev/2026/02/07/Record-Listening-Club-Albums/"><![CDATA[<div style="display: flex; justify-content: space-around; width: fit-content; margin: 0 auto;">
    <img src="/assets/images/bittersweet.jpg" alt="bittersweet" style="width:33%; height: auto; padding: 5px;" />
    <img src="/assets/images/to_be_a_man.png" alt="to be a man" style="width: 33%; height: auto; padding: 5px;" />
    <img src="/assets/images/natchl_blues.jpg" alt="the natch'l blues" style="width:33%; height: auto; padding: 5px;" />
</div>

<hr />

<p>In addition to the above albums, we listened to ATLiens by Outkast. I’ve been really enjoying their song <em>Babylon</em>, which features Andrea Martin, as well as <em>Wheelz of Steel</em>.</p>]]></content><author><name>maxrios</name></author><summary type="html"><![CDATA[]]></summary></entry><entry><title type="html">BPL’s Record Listening Club</title><link href="https://maxrios.dev/2026/01/10/BPL's-Record-Listening-Club/" rel="alternate" type="text/html" title="BPL’s Record Listening Club" /><published>2026-01-10T00:00:00+00:00</published><updated>2026-01-10T00:00:00+00:00</updated><id>https://maxrios.dev/2026/01/10/BPL&apos;s-Record-Listening-Club</id><content type="html" xml:base="https://maxrios.dev/2026/01/10/BPL&apos;s-Record-Listening-Club/"><![CDATA[<table>
  <tbody>
    <tr>
      <td><img src="/assets/images/bpl_vinyl.jpg" alt="image" /></td>
      <td><img src="/assets/images/record_book_list.jpg" alt="image" /></td>
      <td><img src="/assets/images/record_listening_club.jpg" alt="image" /></td>
    </tr>
  </tbody>
</table>

<hr />

<p><img style="float: left; margin-right: 10px" src="/assets/images/water_from_an_ancient_well.jpg" />
I recently attended the Brooklyn Public Library’s Record Listening Club, at the Adams Street location. It’s a wonderful way to expose yourself to musicians you’ve never heard of. I brought <em>Kollage</em> by Bahamadia and played songs like <em>Spontaneity</em>, <em>Rugged Ruff</em>, and <em>3 the Hard Way</em>. During the listening session, I was introduced to <em>Water from an Ancient Well</em> by Abdullah Ibrahim. His South African jazz was calming and fresh.</p>

<p>This new club is welcoming and diverse!</p>]]></content><author><name>maxrios</name></author><summary type="html"><![CDATA[]]></summary></entry><entry><title type="html">AI, Education, and Why I’m Cautiously Optimistic</title><link href="https://maxrios.dev/2025/06/15/AI-Education-and-Why-I'm-Cautiously-Optimistic/" rel="alternate" type="text/html" title="AI, Education, and Why I’m Cautiously Optimistic" /><published>2025-06-15T00:00:00+00:00</published><updated>2025-06-15T00:00:00+00:00</updated><id>https://maxrios.dev/2025/06/15/AI-Education-and-Why-I&apos;m-Cautiously-Optimistic</id><content type="html" xml:base="https://maxrios.dev/2025/06/15/AI-Education-and-Why-I&apos;m-Cautiously-Optimistic/"><![CDATA[<p>Four years ago, I was introduced to the world of plagiarism detection. It was a new challenge for me, but not for Google Classroom, who had released the tool a few years earlier. The complexity behind it was immense, and its application changed my perspective on the utility of assisted grading tools. At the time (and still even now), everything revolved around the content you had access to. How can you verify that something is not-so-original with a limited corpus? We introduced new ways to tackle this problem, specifically through the intra-domain detection of past student essays. This style of plagiarism detection is almost analog to the challenges that we’re seeing today. Is it possible to detect if a sample of writing was created by AI? Is it actually plagiarism if it was? These are just a couple of the fun questions that began to stir the pot.</p>

<p>My interest in large language models (LLMs) really peaked when the quality of them exploded back in 2021. Around that time, I started having conversations with professors and students about potential use cases and concerns with AI. COVID left a lot of teachers unprepared for fully online learning environments, and I could see this leading to similar issues. Over time, my professors and I discussed the potential for plagiarizing work and AI having influence on papers. What I didn’t fully understand was how quickly these problems would evolve. The conversations quickly transitioned from a lack of effort in essays to substanceless essays. Essays were riddled with incorrect information. Areas that students thought were small knowledge gaps turned out to be large enough for issues like these to slip through. The temptation to just complete an assignment was too great.</p>

<p>And this wasn’t limited to students. College admins were also caught making the exact same mistakes. I’ve had many professors gossip about how speeches and emails had no substance and contained incorrect information. It turned out that no one wanted to struggle to complete their tasks.</p>

<p>These conversations and stories are what grew my sense of unease. And everyday, a new tool was being deployed with a questionable level of quality. I knew that my job in plagiarism detection wasn’t going to be the only thing to become more challenging. I worried (and still worry) about the learning outcomes for students that are introduced to these tools at the start of their education journey. I worried about the false sense of understanding these tools are creating, where real knowledge should be. Was this becoming a problem that went beyond academic integrity and policing plagiarism?</p>

<p>In an attempt to better understand the problem, I started attending more of Classroom’s school visits and teacher studies. My prediction was right, teachers were caught off guard. It might be easy to catch AI generated content now, but what about one to two years from then? Every single teacher I spoke to had a fallback method: know your student. Does my student actually write like this? Did they copy and paste everything into the Google Doc? Why is my 9th grader suddenly writing at a college level? Obviously, for an experienced teacher, it was easy to catch.</p>

<p>Moving into 2024, everything quickly progressed. Students and administrators became more reliant on ChatGPT, Gemini, and Claude, and their work reflected it. However, the evidence became more clear: these problems aren’t new, only the technology is. Misinformation has always existed on websites, readily available via a quick Google search. Students have always shared essays with each other. Everyone has gotten incorrect information from a confident friend at some point in their lives. Though, these AI tools’ ability to combine all of these negative characteristics and make it always available was unique. And continuous improvement would only make these challenges more complex.</p>

<p>Despite all the concerns in these conversations, there was an undercurrent of optimism. Teachers saw these emerging tools as an opportunity to optimize the process for creating differentiated assignments and providing accessible, one-on-one tutoring.</p>

<p>When I first encountered differentiated learning in classrooms, it struck me how much teachers were already juggling. American University School of Education’s <em><a href="https://soeonline.american.edu/blog/differentiated-instruction/" target="\_blank">How Differentiated Instruction Supports All Students</a></em> breaks down what teachers need to track: readiness to learn, learning preferences, prior knowledge, languages spoken, and personal interests. That’s a lot of moving pieces for one teacher managing thirty students. The thing is, teachers are already collecting mountains of data through every assignment, every question asked during class, every moment of confusion. But making sense of all that data? That’s where things get overwhelming, and where I started wondering if AI could actually help rather than complicate things.</p>

<p>The conversations I’ve had with teachers about AI tools for differentiated instruction have been cautiously optimistic. They see the potential for these models to spot patterns they might miss: maybe Sarah always struggles with word problems but excels at visual math, or maybe Marcus’s writing improves dramatically when assignments connect to his interests in sports. But here’s what keeps coming up in every discussion: the fear of bias creeping into these AI systems. Teachers know their students in ways that data points can’t capture, so while automated assignment creation based on student profiles sounds promising, it only works if teachers remain the final decision-makers. The goal isn’t to replace that human judgment; it’s to give teachers better tools to act on what they already know about their students.</p>

<p>Beyond classroom instruction, I kept hearing about another pain point: access to tutoring. Students I talk to struggle to find quality tutors who are both available and affordable. Having someone actually work with you and care about your learning requires dedication that’s hard to come by. What caught my attention about AI tutors was their promise of being there whenever students needed help. Sure, you could debate the quality of that help, but just having something available felt like progress.</p>

<p>Then reality started setting in through my conversations with teachers. Consistent availability means nothing if the teaching quality is terrible. Are these AI tutors actually guiding students with helpful hints, or are they just handing over answers? Do they catch the big misunderstandings that trip students up, or do they get lost in irrelevant details? Are their explanations even reliable? I started asking myself whether these AI tutors needed human supervision to be useful, or if students were better off just doing what they’ve always done: googling their questions and hoping for the best.</p>

<p>After all these conversations with teachers and watching how quickly everything has evolved, I keep coming back to the same realization: we’re not dealing with something entirely new here. We’ve always had tools that could either help students learn or help them avoid learning. AI just happens to be really, really good at both.</p>

<p>The teachers who seem to be handling this best aren’t the ones trying to ban AI or the ones letting students use it for everything. They’re the ones asking students to show their work, including their AI interactions. They’re having students fact-check and improve AI output instead of just accepting it. They’re treating AI like what it actually is: a research assistant that sometimes gets things wrong, not a replacement for thinking.</p>

<p>What worries me most isn’t the technology itself, but how we’re rushing to either embrace it completely or reject it entirely. Both approaches miss the point. The question isn’t whether AI will change education – it already has. The question is whether we’ll teach students to think alongside these tools, or let them outsource their thinking entirely. From where I sit, that choice is going to define what learning looks like for the next generation.</p>]]></content><author><name>maxrios</name></author><summary type="html"><![CDATA[Four years ago, I was introduced to the world of plagiarism detection. It was a new challenge for me, but not for Google Classroom, who had released the tool a few years earlier. The complexity behind it was immense, and its application changed my perspective on the utility of assisted grading tools. At the time (and still even now), everything revolved around the content you had access to. How can you verify that something is not-so-original with a limited corpus? We introduced new ways to tackle this problem, specifically through the intra-domain detection of past student essays. This style of plagiarism detection is almost analog to the challenges that we’re seeing today. Is it possible to detect if a sample of writing was created by AI? Is it actually plagiarism if it was? These are just a couple of the fun questions that began to stir the pot.]]></summary></entry><entry><title type="html">A Pragmatic Comparison of User Impersonation and RBAC</title><link href="https://maxrios.dev/2025/04/11/A-Pragmatic-Comparison-of-User-Impersonation-and-RBAC/" rel="alternate" type="text/html" title="A Pragmatic Comparison of User Impersonation and RBAC" /><published>2025-04-11T00:00:00+00:00</published><updated>2025-04-11T00:00:00+00:00</updated><id>https://maxrios.dev/2025/04/11/A-Pragmatic-Comparison-of-User-Impersonation-and-RBAC</id><content type="html" xml:base="https://maxrios.dev/2025/04/11/A-Pragmatic-Comparison-of-User-Impersonation-and-RBAC/"><![CDATA[<p>The ability for a service to operate beyond its authorized controls, effectively performing actions as the user, forms the core of programmatic user impersonation. User impersonation, or just impersonation, is a powerful tool that can be a convenient solution and a complex troublemaker. Let’s try and visualize this:</p>

<p>Maisie and Max are working on a project that requires them to send a PDF back-and-forth to each other. Maisie completes her part and sends it to Max. Max completes his and sends it back. Uh oh! Max made a mistake and needs to get the document back. How can this be done without giving too much control to Maisie or Max? Impersonation. The file service can impersonate Maisie, send the file back to Max, and log it as returned. Impersonation can also be Max requesting a batch of files to be sent and the service performing the individual actions on Max’s behalf.</p>

<p>Maisie and Max continue working, as well as your potential thousands of users. After millions of real and impersonated file sends, how do we distinguish between these actions?</p>

<p>In your own service, providing a basic log such as “job X returned file on behalf of Max” might suffice. However, any resulting actions that aren’t owned by you will interpret these impersonations as real user actions. For example, if my service is using the Dropbox API and I impersonate Maisie, Dropbox will interpret this request as actually coming from Maisie. This is a problem because all of Dropbox’s audit logs will be invalid from the perspective of your user. A convenient solution suddenly results in downstream mistakes. When considering impersonation in your designs, think about the impact it could have on your service and its dependencies’ audit logs.</p>

<p>So how does this impersonation work under the hood, especially concerning credentials? This form of impersonation is always a service that replicates a user’s action by utilizing that user’s data access token (DAT). This impersonation token can have a time to live (TTL) that varies depending on the service’s needs. Typically, you would limit the TTL of the DAT to a small value, like 5 minutes. Queued jobs might run into expiration issues, where you would need to explore longer TTLs. A long lived DAT might be a security concern within your team, so your impersonation job or action should operate within the time constraint of your user’s DAT. Whether your project is limited by a dependencies QPS ceiling and requires a queue, or security concerns don’t allow you to maintain user credentials, there are alternatives, such as a role-based access control (RBAC) system.</p>

<p>RBAC is a security approach used to restrict system access for authorized users based on their roles within an organization. The role your service would take on would be similar to a super user with access to move all files freely. A dependency might require the user to allow increased permissions for this service, but it would avoid confusion on who is performing the move operations. Job X would still be performing actions on behalf of Max, but now the dependency knows that too. It simplifies the potential issues that could arise from your product’s audit logs. However, this shouldn’t take away from the security concerns that an RBAC system introduces. Assigning a service super user access to all of your users’ files is a major risk, and should be limited or avoided all together. From my experience, limiting the data that the service can access is a possible solution to this problem. This might be in the form of creating labels to categorize data, and then specifying permissions based on these labels. In a sense, creating new, non-super user roles.</p>

<p>Navigating the choice between user impersonation and RBAC often comes down to which set of problems you’d rather solve. Impersonation’s downstream audit ambiguity and token lifecycle management can be complex, while RBAC’s potential for overly broad permissions presents serious security hurdles, often necessitating the kind of granular role definition discussed from my own experience. Both paths require careful implementation. Before committing, thoroughly analyze the impact on your audit logs, your dependencies, and your overall security architecture to determine which approach — or perhaps a carefully architected hybrid — best aligns with your project’s specific constraints and priorities.</p>]]></content><author><name>maxrios</name></author><summary type="html"><![CDATA[The ability for a service to operate beyond its authorized controls, effectively performing actions as the user, forms the core of programmatic user impersonation. User impersonation, or just impersonation, is a powerful tool that can be a convenient solution and a complex troublemaker. Let’s try and visualize this:]]></summary></entry><entry><title type="html">Introducing Moody</title><link href="https://maxrios.dev/2025/02/06/Introducing-Moody/" rel="alternate" type="text/html" title="Introducing Moody" /><published>2025-02-06T00:00:00+00:00</published><updated>2025-02-06T00:00:00+00:00</updated><id>https://maxrios.dev/2025/02/06/Introducing-Moody</id><content type="html" xml:base="https://maxrios.dev/2025/02/06/Introducing-Moody/"><![CDATA[<p>Journaling, or keeping a diary, is a great way to self-reflect. It’s an important part of my life. Over the years I’ve found different ways to journal. I’ve written in notebooks, I’ve written on sticky notes, I’ve written in note apps. It doesn’t really matter where you write, it just matters that you think about what’s trapped in your mind.</p>

<p>In 2024, I was introduced to mood tracking apps. A silly little icon that’s typically linked with a journal entry of how you feel in the moment or the day. My favorite to use was <a href="https://play.google.com/store/apps/details?id=com.marchlab.haema&amp;hl=en_US">Haema</a>, a minimal-style mood tracking app that had a small feature set. I mostly liked it because of a basic calendar view of all my moods and a really cool widget. After using it for a few weeks, I was disapointed to learn that the payment feature, that allowed you to continue using the app after the free trial, was broken. No more Haema, not more simply designed, worry-free mood tracking. And that was my lightbulb moment. Why not just take all of my favorite features from the mood tracking apps I’ve tried and make one myself? A fun little side project.</p>

<p><em>Spolier Alert: It was not little.</em></p>

<p>Welcome, Moody!</p>

<p>Moody is a diary and mood tracking app I’ve been developing since September of 2024. Its goal is to help people better understand their emotional patterns through simple, daily mood tracking – making self-reflection an accessible and sustainable habit rather than another task on their to-do list.</p>

<p>And to me, everything free is a major cornerstone of this goal. Enjoy all the features without worrying about a price tag.</p>

<p>It’s currently being tested in a closed pilot program on iOS and Android. If you’re interested in trying it out, join the <a href="https://forms.gle/i5y6mFbkiY87zBVF9">early access group</a>!</p>

<p>I plan on writing a bit more about Moody in the future, specifically on my experience using Flutter.</p>

<p>Stay tuned!</p>]]></content><author><name>maxrios</name></author><summary type="html"><![CDATA[Journaling, or keeping a diary, is a great way to self-reflect. It’s an important part of my life. Over the years I’ve found different ways to journal. I’ve written in notebooks, I’ve written on sticky notes, I’ve written in note apps. It doesn’t really matter where you write, it just matters that you think about what’s trapped in your mind.]]></summary></entry><entry><title type="html">Welcome!</title><link href="https://maxrios.dev/2025/01/19/Welcome/" rel="alternate" type="text/html" title="Welcome!" /><published>2025-01-19T00:00:00+00:00</published><updated>2025-01-19T00:00:00+00:00</updated><id>https://maxrios.dev/2025/01/19/Welcome</id><content type="html" xml:base="https://maxrios.dev/2025/01/19/Welcome/"><![CDATA[<p><em>Welcome inside of Max’s mind…</em></p>

<p><br />
A blog / log / notebook / brain dump on tech, video games, books, music, or whatever. Expect a mixed bag of posts that will likely never have a schedule. Some might even just be pictures. The goal is to consolidate my thoughts recorded in far too many notebooks (they’re just so fun to write in) and some Google Docs.</p>

<p>I think a lot goes on in a day, and days are far too long to store everything in your head. As a person with an <em>awful</em> memory, I found these little (occasionally schmedium) notes as a nice reminder of what actually happened. Sometimes a note in a mood tracker might suffice. Sometimes thoughts just need to be forgotten.</p>

<p>Unfortunately, I won’t be spending much energy editing these entries. Damare, I apologize in advance. Everyone else, feel free to voice your concerns to the mid-night air – a surprisingly rare occurence.</p>

<p>xoxo</p>

<p><img src="/assets/images/dog.png" alt="image" /></p>]]></content><author><name>maxrios</name></author><summary type="html"><![CDATA[Welcome inside of Max’s mind…]]></summary></entry></feed>