<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Zainab Mosunmola on Medium]]></title>
        <description><![CDATA[Stories by Zainab Mosunmola on Medium]]></description>
        <link>https://medium.com/@zainabmosunmola?source=rss-334bc9672bd4------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*dJFl2IOV3-eDHUGO6s5FhQ.jpeg</url>
            <title>Stories by Zainab Mosunmola on Medium</title>
            <link>https://medium.com/@zainabmosunmola?source=rss-334bc9672bd4------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 05:08:01 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@zainabmosunmola/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[I’m Building a Real-Time Data Engineering Platform (And I Have No Idea What I’m Doing)]]></title>
            <link>https://zainabmosunmola.medium.com/day-1-im-building-a-real-time-data-engineering-platform-and-i-have-no-idea-what-i-m-doing-4a85bdb6556e?source=rss-334bc9672bd4------2</link>
            <guid isPermaLink="false">https://medium.com/p/4a85bdb6556e</guid>
            <category><![CDATA[data-engineering]]></category>
            <category><![CDATA[docker]]></category>
            <category><![CDATA[pipeline]]></category>
            <category><![CDATA[kafka]]></category>
            <category><![CDATA[airflow]]></category>
            <dc:creator><![CDATA[Zainab Mosunmola]]></dc:creator>
            <pubDate>Thu, 29 Jan 2026 05:45:04 GMT</pubDate>
            <atom:updated>2026-01-29T05:45:25.889Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A diary of building a full-scale data engineering project from scratch</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/0*gfMSut_rNk-cpiEe.jpg" /><figcaption>source: Pinterest</figcaption></figure><p>Let me even be honest with you.</p><p>I’ve done the tutorials. I’ve followed the courses. I am always learning something. Starting. Not Finishing. I feel like I have been stuck in a learning loop for a long time. But I’ve never built a <strong>real</strong> project from scratch.</p><p>Today, that changes, I guess.</p><p>I’ll document everything. The wins. The failures. The moments where I stare at my screen, wondering what I’ve gotten myself into.</p><p>So, this is Day 1.</p><h3>The Project: Manitoba Grid Intelligence</h3><p>I’m building a real-time power grid analytics platform. Why power grids? Because:</p><ol><li>Manitoba Hydro is one of the largest utilities in North America</li><li>Power grid data is time-series data</li><li>It combines streaming + batch processing (the full picture)</li><li>I have an engineering background, and I love engineering duhhhh</li></ol><p>The platform will:</p><ul><li>Ingest simulated sensor data in real-time</li><li>Process weather data from Environment Canada</li><li>Detect anomalies in equipment readings</li><li>Score equipment health</li><li>Forecast electricity demand</li></ul><p>Ambitious? Yes. Will I finish? We shall see.</p><h3>The Architecture</h3><p>Here’s what I’m building:</p><pre>┌─────────────────────────────────────────────────────────────────┐<br>│               MANITOBA GRID INTELLIGENCE                        │<br>├─────────────────────────────────────────────────────────────────┤<br>│                                                                 │<br>│  DATA SOURCES                                                   │<br>│  ────────────                                                   │<br>│  • Simulated SCADA sensors (real-time)                          │<br>│  • Environment Canada API (weather)                             │<br>│  • MISO market prices (batch)                                   │<br>│                                                                 │<br>│                          │                                      │<br>│                          ▼                                      │<br>│                                                                 │<br>│  STREAMING LAYER: Kafka                                         │<br>│  ──────────────────────                                         │<br>│  • Buffers incoming data                                        │<br>│  • Decouples producers from consumers                           │<br>│  • Handles traffic spikes                                       │<br>│                                                                 │<br>│                          │                                      │<br>│                          ▼                                      │<br>│                                                                 │<br>│  PROCESSING + STORAGE: TimescaleDB                              │<br>│  ─────────────────────────────────                              │<br>│  • Time-series optimized PostgreSQL                             │<br>│  • Bronze/Silver/Gold data layers                               │<br>│  • Hypertables for efficient queries                            │<br>│                                                                 │<br>│                          │                                      │<br>│                          ▼                                      │<br>│                                                                 │<br>│  ORCHESTRATION: Airflow                                         │<br>│  ──────────────────────                                         │<br>│  • Schedules batch jobs                                         │<br>│  • Manages dependencies                                         │<br>│  • Monitors pipeline health                                     │<br>│                                                                 │<br>└─────────────────────────────────────────────────────────────────┘</pre><h3>My Docker Compose:</h3><p>Let me break down every tool in my stack and why it’s there.</p><h3>1. Apache Kafka: The Message Broker</h3><p><strong>What it does: </strong>A distributed streaming platform. Catches real-time data and holds it until something is ready to process it.</p><p><strong>Why I need it:</strong> Imagine sensors sending 1000 readings per second. Without Kafka, if my database is slow for 2 seconds, I lose 2000 readings. With Kafka, data waits in line. Nothing lost.</p><p><strong>Analogy:</strong> Kafka is like a post office. Senders drop off packages (messages), receivers pick them up when ready. The post office doesn’t care if the receiver is home.</p><h4>Where it fits:</h4><pre>┌──────────────────────────────────────────────────────────────┐<br>│                                                              │<br>│  [Sensors] ──produce──▶ [KAFKA] ──consume──▶ [Processing]   │<br>│                           │                                  │<br>│                    Message Buffer                            │<br>│                    - Holds data                              │<br>│                    - Handles spikes                          │<br>│                    - Enables replay                          │<br>│                                                              │<br>└──────────────────────────────────────────────────────────────┘</pre><h3>2. Apache Zookeeper: The Coordinator</h3><p><strong>What it does:</strong> Manages Kafka’s internal state. Tracks which Kafka brokers are alive, who leads which partition.</p><p><strong>Why I need it:</strong> Kafka requires it (in older versions). Think of Zookeeper as the manager who keeps track of everyone’s shifts.</p><h4>Note:</h4><p>Newer Kafka (3.x+) can run without Zookeeper using “KRaft mode”. But most tutorials, courses, and Docker images still use Zookeeper. So I’m learning the traditional way, I guess.</p><h3>3. TimescaleDB: The Time-Series Database</h3><p><strong>What it does:</strong> PostgreSQL with superpowers for time-series data.</p><p><strong>Why I need it:</strong> Regular PostgreSQL slows down when you have millions of sensor readings. TimescaleDB automatically partitions data by time, compresses old data, and optimizes time-based queries.</p><p><strong>Why not InfluxDB?</strong> TimescaleDB uses standard SQL. I already know SQL. Learning a new query language isn’t the goal here *side eye*.</p><h4>Key config explained:</h4><ul><li>ports: &quot;5432:5432&quot; - I can connect from my laptop at localhost:5432</li><li>volumes: postgres-data - Data persists even if container restarts</li><li>init-db.sql - Runs on first startup to create my schemas</li><li>healthcheck - Docker checks if database is actually responding</li></ul><h3>4. Apache Airflow: The Orchestrator</h3><p><strong>What it does: </strong>A workflow orchestration platform. It schedules, runs, and monitors batch jobs.</p><p><strong>Why I need it:</strong> Some data comes in batches (weather forecasts, daily prices). I need something to:</p><ul><li>Run jobs on schedule (every hour, every day)</li><li>Handle failures and retries</li><li>Track what ran and when</li></ul><p><strong>Analogy:</strong> Airflow is like a cron job on steroids with a beautiful UI.</p><h4>4.1: Airflow Webserver</h4><p>The UI component. Let’s you:</p><ul><li>See all your workflows (DAGs)</li><li>Trigger jobs manually</li><li>View logs</li><li>Monitor task status</li></ul><h4>Why do I need it?</h4><p>Some data comes in batches, not streams:</p><ul><li>Weather forecasts (every hour)</li><li>Market prices (every 5 minutes)</li><li>Daily reports (once a day)</li></ul><p>Airflow handles: “Run this job every hour. If it fails, retry 3 times. Email me if it still fails.”</p><h4>Key config explained:</h4><ul><li>build: dockerfile - Uses custom Dockerfile to add my Python packages</li><li>EXECUTOR: LocalExecutor - Runs multiple tasks in parallel on one machine</li><li>LOAD_EXAMPLES: false - Don&#39;t show tutorial DAGs (keeps UI clean)</li><li>volumes: ./dags - My DAG files. Edit locally, Airflow sees changes immediately. Defines the workflow.</li></ul><h4>4.2: Airflow Scheduler</h4><p>The brain of Airflow. It:</p><ul><li>Monitors all DAG files</li><li>Decides when to run tasks</li><li>Queues tasks for execution</li><li>Tracks dependencies between tasks</li></ul><blockquote>Scheduler does the work, Webserver lets you see what’s happening.</blockquote><h4>Why separate containers?</h4><p>Separation of concerns. If the UI crashes, jobs still run. If the scheduler crashes, you can still view logs.</p><h4>4.3: Airflow Init</h4><p>A one-time setup container that:</p><ol><li>Creates Airflow’s database tables (airflow db init)</li><li>Creates an admin user so I can log in</li></ol><h4>Why a separate container?</h4><p>It runs once and exits. Web server and Scheduler run forever. Different lifecycles, different containers.</p><h3>5. Adminer: The Database UI</h3><p><strong>What it does: </strong>A lightweight database management UI. Like pgAdmin but simpler.</p><p>Essentially, a web interface to browse my database.</p><p><strong>Why I need it:</strong> So I can see what’s actually in my tables without writing SQL every time.</p><h3>How They All Connect</h3><pre>┌─────────────────────────────────────────────────────────────────────────────────┐<br>│                         COMPLETE SYSTEM                                          │<br>├─────────────────────────────────────────────────────────────────────────────────┤<br>│                                                                                 │<br>│  MY LAPTOP                                                                      │<br>│  ───────────                                                                    │<br>│                                                                                 │<br>│  localhost:8081 ──▶ Airflow UI      (see workflows)                             │<br>│  localhost:8080 ──▶ Adminer         (browse database)                           │<br>│  localhost:9094 ──▶ Kafka           (send messages)                             │<br>│  localhost:5432 ──▶ TimescaleDB     (SQL queries)                               │<br>│                                                                                 │<br>├─────────────────────────────────────────────────────────────────────────────────┤<br>│                                                                                 │<br>│  INSIDE DOCKER (grid-network)                                                   │<br>│  ────────────────────────────                                                   │<br>│                                                                                 │<br>│                         ┌───────────────┐                                       │<br>│                         │   Zookeeper   │                                       │<br>│                         │    :2181      │                                       │<br>│                         └───────┬───────┘                                       │<br>│                                 │ coordinates                                   │<br>│                         ┌───────▼───────┐                                       │<br>│   Sensors ─────────────▶│     Kafka     │                                       │<br>│                         │  :9092/:9094  │                                       │<br>│                         └───────┬───────┘                                       │<br>│                                 │                                               │<br>│                                 │ data flows to                                 │<br>│                                 ▼                                               │<br>│   ┌─────────────────────────────────────────────────────────────────┐          │<br>│   │                      TimescaleDB                                │          │<br>│   │                        :5432                                    │          │<br>│   │                                                                 │          │<br>│   │   Stores:                                                       │          │<br>│   │   • Sensor readings (hypertable)                                │          │<br>│   │   • Weather data                                                │          │<br>│   │   • Market prices                                               │          │<br>│   │   • Airflow metadata                                            │          │<br>│   │                                                                 │          │<br>│   └─────────────────────────────────────────────────────────────────┘          │<br>│                                 │                                               │<br>│                                 │ queries                                       │<br>│                                 ▼                                               │<br>│   ┌─────────────────────────────────────────────────────────────────┐          │<br>│   │                        AIRFLOW                                  │          │<br>│   │                                                                 │          │<br>│   │   ┌───────────┐    ┌───────────┐    ┌───────────┐              │          │<br>│   │   │  Init     │    │ Scheduler │    │ Webserver │              │          │<br>│   │   │(one-time) │    │  (brain)  │    │   (UI)    │              │          │<br>│   │   └───────────┘    └───────────┘    └───────────┘              │          │<br>│   │                                                                 │          │<br>│   │   Runs batch jobs:                                              │          │<br>│   │   • Fetch weather hourly                                        │          │<br>│   │   • Transform data daily                                        │          │<br>│   │   • Train ML model                                              |          │          <br>│   └─────────────────────────────────────────────────────────────────┘          │<br>│                                                                                │<br>│   ┌───────────┐                                                                │<br>│   │  Adminer  │◀──── ME (browse tables, run queries)                           │<br>│   │   :8080   │                                                                │<br>│   └───────────┘                                                                │<br>│                                                                                 │<br>└─────────────────────────────────────────────────────────────────────────────────┘</pre><blockquote>So, this one YAML file that defines my entire infrastructure</blockquote><h3>Did I Write This YAML From Scratch?</h3><p>Lol no.</p><p>I told Claude Code: “Create a Docker Compose with TimescaleDB, Kafka, Airflow, and Adminer for my data engineering project.”</p><p>It generated everything.</p><p>I generated the architecture and tools lol</p><p>Human-in-the-loop 😂</p><h3>Day 1 Progress</h3><p><strong>What I accomplished:</strong></p><ul><li>[x] Project scaffold created</li><li>[x] Docker Compose file with all services</li><li>[x] Understanding of each tool’s purpose</li><li>[x] Currently downloading ~3GB of Docker images</li></ul><p><strong>What I learned:</strong></p><ul><li>Kafka needs Zookeeper (for now)</li><li>TimescaleDB is just PostgreSQL with extensions</li><li>Airflow has 3 components: webserver, scheduler, and init</li><li>YAML anchors let you reuse configuration</li></ul><p><strong>What confused me:</strong></p><ul><li>Kafka’s networking (why two ports??)</li><li>Airflow’s executor types</li><li>Why Docker takes SO LONG the first time, I feel like I am downloading the whole internet chale😭</li></ul><h3>What’s Next</h3><p>Tomorrow I’ll:</p><ol><li>I’ll probably get my container running</li><li>Or contain a disaster, we shall see🙈😂</li></ol><h3>Follow Along</h3><ul><li><a href="https://github.com/mosunmola-zainab/Manitoba-Grid-Intelligence"><strong>GitHub</strong></a></li><li><a href="https://www.linkedin.com/in/zainab-mosunmola-hassan-780b54175/"><strong>LinkedIn</strong></a></li></ul><h3>The Honest Truth</h3><p>I’m writing this while watching Docker download images. I don’t fully understand everything yet. Half of what I wrote above, I learned TODAY tbh.</p><p>See you tomorrow.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4a85bdb6556e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding Terraform’s Fundamentals]]></title>
            <link>https://zainabmosunmola.medium.com/understanding-terraforms-fundamentals-a0eac0d8c9ed?source=rss-334bc9672bd4------2</link>
            <guid isPermaLink="false">https://medium.com/p/a0eac0d8c9ed</guid>
            <category><![CDATA[data-engineering]]></category>
            <category><![CDATA[cloud]]></category>
            <category><![CDATA[terraform]]></category>
            <category><![CDATA[infrastructure-as-code]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <dc:creator><![CDATA[Zainab Mosunmola]]></dc:creator>
            <pubDate>Wed, 28 Jan 2026 16:26:01 GMT</pubDate>
            <atom:updated>2026-01-28T16:26:01.846Z</atom:updated>
            <content:encoded><![CDATA[<h4>Its core concepts, structure, and workflow</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/736/0*s-xcHTWP-SPA_OMH.jpg" /><figcaption>source: Pinterest</figcaption></figure><p>Previously, I wrote about Infrastructure as Code (IaC), and I spoke a little about Terraform there. If we played a little game and asked everyone what tool comes to mind when they hear IaC, almost everyone would probably say Terraform.</p><p>So what makes Terraform the diva?</p><h4>Let’s start with a disclaimer:</h4><p><strong>Terraform is IaC, but not all IaC is Terraform.</strong></p><p>Think of it like this:</p><p>IaC is programming (the concept)</p><p>Terraform is Python (the tool)</p><p>So why Terraform?</p><ol><li><strong>Multi-cloud</strong></li></ol><ul><li>It works with multiple cloud platforms, AWS, Google Cloud, Azure, Kubernetes</li><li>Same syntax for all clouds!</li><li>You have a wide palate, you’re not locked into one provider.</li></ul><p><strong>2. Declarative</strong></p><p>You have a wide palate, and you want to order whatever you want without having to say how to build it. Terraform it is!</p><blockquote>You say <strong>WHAT</strong> you want, not <strong>HOW</strong> to build it:</blockquote><p>Terraform figures out the steps, leave the cooking to it!</p><p><strong>3. State Management</strong></p><p>This is one of the most important concepts in IaC.</p><p>If Terraform is the Architect, the State File is the Current Blueprint that tells the architect exactly what has already been built.</p><p>If Terraform is the Chef, the State File is the recipe book that tracks and tells you what exactly is already cooking.</p><p>So essentially, it remembers what has been built, so it knows what changes and gets deleted later</p><h3>The Terraform Blueprint</h3><p>The Core Files:</p><h4>1. main.tf - The Main Blueprint</h4><p>This one contains your actual infrastructure definitions.</p><pre>resource &quot;google_storage_bucket&quot; &quot;data_lake&quot; {<br>  name     = &quot;my-data-lake&quot;<br>  location = &quot;US&quot;<br>}</pre><pre>resource &quot;google_bigquery_dataset&quot; &quot;my_dataset&quot; {<br>  dataset_id = &quot;ny_taxi&quot;<br>  location   = &quot;US&quot;<br>}</pre><p>This is your construction blueprint</p><h4>2. variables.tf - The Settings</h4><p>These ones are inputs you can change without touching main.tf</p><pre>variable &quot;project_id&quot; {<br>  description = &quot;Your GCP project ID&quot;<br>  type        = string<br>}</pre><pre>variable &quot;region&quot; {<br>  description = &quot;GCP region&quot;<br>  default     = &quot;us-central1&quot;<br>}</pre><p>This is your configuration setting.</p><p><strong>Why would you need one?</strong></p><p>…because you would probably use the SAME code for dev/staging/production by just changing variables!</p><h4>3. .tfstate - The Memory</h4><p>This one is very important. This is the STATE file! It tracks what Terraform has built.</p><pre>{<br>  &quot;resources&quot;: [<br>    {<br>      &quot;type&quot;: &quot;google_storage_bucket&quot;,<br>      &quot;name&quot;: &quot;data_lake&quot;,<br>      &quot;id&quot;: &quot;my-data-lake-12345&quot;<br>    }<br>  ]<br>}</pre><p>If you lose this file, Terraform forgets what it built! Understood? Huh? Okay, good!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*EGgzP5Kc6yYIXi3b.gif" /></figure><h4>4. output.tf - The Results</h4><p>This one displays information after creation.</p><p>It is the part of the code that tells Terraform: “Once you are done building everything, please tell me so and so specific details so I don’t have to go looking for them”.</p><pre>output &quot;bucket_name&quot; {<br>  value = google_storage_bucket.data_lake.name<br>}</pre><p>Your receipt after checkout.</p><h3>The Building Blocks</h3><h4>1. terraform Block - The Settings</h4><pre>terraform {<br>  required_version = &quot;&gt;= 1.0&quot;  # Minimum Terraform version needed<br>  <br>  backend &quot;local&quot; {  # Where to store the .tfstate file<br>    path = &quot;terraform.tfstate&quot;<br>  }<br>  <br>  required_providers {  # Which cloud providers you&#39;re using<br>    google = {<br>      source  = &quot;hashicorp/google&quot;<br>      version = &quot;~&gt; 4.0&quot;<br>    }<br>  }<br>}</pre><p><strong>Translation:</strong></p><ul><li>“I need Terraform version 1.0 or higher”</li><li>“Store my state file locally”</li><li>“I want to use Google Cloud”</li></ul><h4>2. provider Block - The Cloud Connection</h4><pre>provider &quot;google&quot; {<br>  project = &quot;my-project-id&quot;<br>  region  = &quot;us-central1&quot;<br>}</pre><p><strong>Translation:</strong> “Connect to Google Cloud using this project and region”</p><p>Those are your login credentials to the cloud</p><h4>3. resource Block - The Actual Infrastructure</h4><p>This is where you specify what you want to build in the cloud.</p><pre>resource &quot;google_storage_bucket&quot; &quot;my_bucket&quot; {<br>  name     = &quot;unique-bucket-name&quot;<br>  location = &quot;US&quot;<br>}<br>```</pre><pre>**Anatomy:**<br>```<br>resource &quot;TYPE&quot; &quot;NAME&quot; {<br>  ↑         ↑       ↑<br>  |         |       └─ Your custom name (for Terraform to reference)<br>  |         └───────── Cloud provider&#39;s resource type<br>  └─────────────────── Keyword<br>  <br>  key = value  ← Configuration<br>}</pre><h4>4. variable &amp; locals - Dynamic Values</h4><p>This is what makes your code flexible, you don’t have to repeat yourself a million times. It works like a FUNCTION.</p><p><strong>Variables (User Input):</strong></p><pre>variable &quot;bucket_name&quot; {<br>  type    = string<br>  default = &quot;my-bucket&quot;<br>}</pre><pre># Use it:<br>resource &quot;google_storage_bucket&quot; &quot;bucket&quot; {<br>  name = var.bucket_name  # ← References the variable<br>}</pre><p><strong>Locals (Constants/Computed Values):</strong></p><pre>locals {<br>  full_bucket_name = &quot;${var.project_id}-${var.bucket_name}&quot;<br>}</pre><pre># Use it:<br>resource &quot;google_storage_bucket&quot; &quot;bucket&quot; {<br>  name = local.full_bucket_name<br>}</pre><h3>The Workflow</h3><p>How do we actually use it?</p><p><strong>The 4 Commands You Need:</strong></p><h4>1. terraform init - Setup</h4><pre>terraform init</pre><p><strong>What it does:</strong></p><ul><li>Initializes terraform</li><li>Downloads the Google Cloud provider plugin</li><li>Sets up the backend (where to store state)</li><li>Prepares your working directory</li></ul><p><strong>When to run:</strong> FIRST TIME, or when you add new providers</p><p>Think of it as installing dependencies (like npm install, pip install)</p><h4>2. terraform plan - Preview</h4><pre>terraform plan</pre><p>What it does:</p><ul><li>Compares your .tf files to the current cloud state</li><li>Shows what will be created/changed/destroyed</li><li>DOES NOT make changes!</li></ul><p>The best part about it is that it works as a <strong>Dry Run</strong>. It doesn&#39;t actually touch Cloud yet. It just gives you a summary/preview of what the code would be doing.</p><p>You want to run something like this:</p><pre># google_storage_bucket.data_lake will be created<br>  + resource &quot;google_storage_bucket&quot; &quot;data_lake&quot; {<br>      + name     = &quot;my-data-lake&quot;<br>      + location = &quot;US&quot;<br>    }</pre><p>It outputs something like this:</p><p>Plan: 1 to add, 0 to change, 0 to destroy.</p><p>For example, if you see “1 to destroy” and you didn’t mean to delete anything, you can stop, fix your code, and plan again.</p><h4>3. terraform apply - Execute</h4><pre>terraform apply</pre><p><strong>What it does:</strong></p><ol><li>Runs terraform plan automatically</li><li>Shows you the plan</li><li>Asks: “Do you want to proceed? (yes/no)”</li><li><strong>This is what creates the actual infrastructure in the cloud!</strong></li><li>Updates .tfstate with what was created</li></ol><h4>4. terraform destroy - Cleanup</h4><pre>terraform destroy</pre><p><strong>What it does:</strong></p><ul><li><strong>DELETES everything</strong> Terraform created</li><li>Asks for confirmation first though… “Are you sure, girl?”</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*HSVUM8VS_3B1WFFB.gif" /></figure><ul><li>Then it updates .tfstate to empty</li></ul><p>This is PERMANENT! Use it carefully! Do you understand? Huh? Okay good!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*PyfoH8hC2QnqX45X.gif" /></figure><h3>PUTTING IT ALL TOGETHER</h3><h4>Example of a simple Workflow:</h4><p><strong>Step 1: Write your infrastructure code</strong></p><p><strong>main.tf:</strong></p><pre>terraform {<br>  required_providers {<br>    google = {<br>      source = &quot;hashicorp/google&quot;<br>    }<br>  }<br>}</pre><pre>provider &quot;google&quot; {<br>  project = &quot;my-project-123&quot;<br>  region  = &quot;us-central1&quot;<br>}</pre><pre>resource &quot;google_storage_bucket&quot; &quot;data_lake&quot; {<br>  name     = &quot;my-data-lake-bucket&quot;<br>  location = &quot;US&quot;<br>}</pre><p><strong>Step 2: Initialize</strong></p><pre>terraform init</pre><p>It shows you if it is successful or not</p><p><strong>Step 3: Preview</strong></p><pre>terraform plan</pre><p>Shows you what it plans to do, literally, in this case, create a storage bucket.</p><p><strong>Step 4: Create</strong></p><pre>terraform apply</pre><p>Type: yes ? Shows you success and what it did.</p><p><strong>Step 5: Verify</strong></p><p>Go to Google Cloud Console → You see your bucket!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*2cAM0yV-oFy4cqQT.gif" /></figure><h3>Step 6: Cleanup (when done)</h3><pre>terraform destroy<br>```<br>Type: `yes`<br>Output: &quot;Destroy complete! Resources: 1 destroyed&quot;</pre><h3>Terraform vs Docker Compose</h3><h4>What it manages:</h4><p>Docker Compose: containers on your computer</p><p>Terraform: cloud infrastructure</p><h4>Configuration file:</h4><p>Docker Compose: docker-compose.yaml</p><p>Terraform: main.tf</p><blockquote>Both are IaC, Just different scopes</blockquote><p>Docker packages your data pipelines.</p><p>Terraform builds the cloud environment they run on.</p><p>Together, they form the backbone of modern data platforms…reproducible, scalable, and automated.</p><p>This is how we move from local notebooks to real production systems.</p><p>Thank you for coming to my TEDTALK!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a0eac0d8c9ed" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is Infrastructure as Code (IaC) and How Does It Work?]]></title>
            <link>https://zainabmosunmola.medium.com/what-is-infrastructure-as-code-iac-and-how-does-it-work-41c89a72b8d9?source=rss-334bc9672bd4------2</link>
            <guid isPermaLink="false">https://medium.com/p/41c89a72b8d9</guid>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[docker]]></category>
            <category><![CDATA[data-engineering]]></category>
            <category><![CDATA[infrastructure-as-code]]></category>
            <category><![CDATA[terraform]]></category>
            <dc:creator><![CDATA[Zainab Mosunmola]]></dc:creator>
            <pubDate>Wed, 28 Jan 2026 05:14:18 GMT</pubDate>
            <atom:updated>2026-01-28T05:48:38.535Z</atom:updated>
            <content:encoded><![CDATA[<p>Do you understand containers? Great! You are one step away from constructing cloud platforms with code!!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/736/0*j3C3GJsGMtNobykh.jpg" /><figcaption>source: Pinterest</figcaption></figure><p>I made a joke recently on LinkedIn about how all the pipeline talk has me picturing oil pipelines and containers floating in the high sea.</p><p>Now, Infrastructure as Code me imagining myself constructing oil rigs. 😂</p><p>If you don’t find it funny 😭, you can forget the joke and jump right into what you came here for in the first place.😒</p><h3>THE FUNDAMENTALS</h3><p><strong>What is Infrastructure?</strong></p><p>Before we even start talking about Infrastructure as Code, let’s talk about what Infrastructure even means.</p><p><strong>Infrastructure </strong>in programming means the foundation that runs your applications.</p><p><strong>Infrastructure as Code</strong> defines the cloud resources that run that code.</p><h3>THE EVOLUTION</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/0*bgSzt5bB-BcBmaCb.gif" /></figure><p>Let’s do a quick history lesson, shall we?</p><h4>The Era of On-Prem Infrastructure (1990s — 2000s)</h4><p><strong>How it worked:</strong></p><ul><li>Buy physical server ($10,000+)</li><li>Rack it in data centers</li><li>Install operating system from CD</li><li>Configure network by hand</li><li>Install software manually</li><li>Document (maybe) in Word doc</li></ul><p>But there were problems, obviously, there would be problems!</p><ul><li>It would take time, we are talking weeks, maybe even months</li><li>Human errors</li><li>Can’t reproduce easily</li><li>Documentation gets outdated</li><li>“It works on my machine, I can’t figure it out on mine”.</li></ul><h4>Infrastructure as a Service (2006–2010s)</h4><p>The Cloud Revolution began. AWS launched and said, “Come rent servers by the hour.”</p><p><strong>How did it work?</strong></p><ul><li>Log in to the AWS console (web interface)</li><li>Click “Launch Instance”</li><li>Fill out forms</li><li>Click, click, click…</li><li>Server is ready in minutes!</li></ul><p>Better, but there were still problems, chale! (And yes, plenty of them!)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*jyTUPRN5DxNCaOe3.gif" /></figure><ul><li>Still manual clicking</li><li>You can’t version control clicks</li><li>Still hard to reproduce, but at least better than the previous era</li><li>No audit trail</li></ul><h4>Infrastructure as Code (2010s — Present)</h4><p>The breakthrough came!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/0*yKHZ-wAVRHi0JqnP.gif" /></figure><p>What if we could just write the code to create infrastructure, just like we were already writing it for applications?</p><pre>resource &quot;aws_instance&quot; &quot;web_server&quot; {<br> ami = &quot;ami-12345&quot;<br> instance_type = &quot;t2.micro&quot;<br>}</pre><p>Then run terraform apply</p><p>Server appears!</p><p>Everyone is happy!</p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/0*yKHZ-wAVRHi0JqnP.gif" /></figure><h3>CORE PRINCIPLES OF IaC</h3><h4>Principle 1: Declarative &gt; Imperative</h4><p><strong>Imperative (How to do it):</strong></p><p>You tell the chef how to cook your food; you give the chef your recipe.</p><p># Step-by-step instructions<br>1. Create VM<br>2. Wait for VM to be ready<br>3. Install software<br>4. Configure firewall<br>5. If error, rollback</p><p><strong>Declarative (What you want):</strong></p><pre># Just describe the end result<br>resource &quot;vm&quot; {<br>  name = &quot;web-server&quot;<br>  size = &quot;small&quot;<br>  software = [&quot;nginx&quot;, &quot;python&quot;]<br>}</pre><p>You don’t tell the chef how to cook it, <strong>YOU JUST LET HIM COOK!</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Fc1DWNX5v47e1ZP5.gif" /></figure><p><strong>Why is declarative better?</strong></p><ul><li>The tool figures out the steps</li><li>Same result every time</li><li>Tool handles errors/retries</li><li>Easier to read and understand</li></ul><h4>Principle 2: Version Control Everything</h4><p>Your infrastructure code lives in Git!</p><pre>git commit -m &quot;Add production database&quot;<br>git commit -m &quot;Increase server size&quot;<br>git commit -m &quot;Add backup storage&quot;</pre><ul><li>You see who changed what and when</li><li>You can rollback bad changes</li><li>Review changes before applying them</li><li>Collaborate with team</li><li>Audit trail for compliance.</li></ul><h4>Principle 3: Idempotent Operations</h4><p>Fancy word eh…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*f6P3TviTWNY-_zjx.gif" /></figure><p>Idempotent means running the same command multiple times, but it produces the same result</p><p><strong>What I mean:</strong></p><p>First run:</p><pre>terraform apply<br># Creates: 1 server, 1 database, 1 storage bucket</pre><p>Second run:</p><pre>terraform apply<br># Output: &quot;No changes. Infrastructure matches your code.&quot;</pre><p><strong>You can literally run it repeatedly!</strong></p><p><strong>Non-idempotent</strong></p><pre>create_server()  # Creates server<br>create_server()  # Creates ANOTHER server (oops!)</pre><h4>Principle 4: Immutable Infrastructure</h4><p>Immutable means you can’t change it, you just replace it.</p><p><strong>Old Way (Mutable):</strong></p><p>Server exists → SSH in → Install updates → Server modified</p><p>Problem? Server “drifts” from original state</p><p><strong>IaC Way (Immutable):</strong></p><p>Server v1 exists → Create server v2 with updates → Delete server v1</p><p>Result: Always know the exact state</p><p><strong>Just like Docker containers! You don’t update or modify a running container, you replace it!</strong></p><h4>Principle 5: Self-Documenting</h4><p><strong>Your code is literally the documentation!</strong></p><p>Instead of:</p><pre>Word Doc: &quot;We have 3 servers in us-east-1...&quot;<br>Reality: Actually 5 servers, doc is outdated</pre><p>With IaC:</p><pre># Your code is the source of truth<br>resource &quot;aws_instance&quot; &quot;web&quot; {<br>  count = 3<br>  region = &quot;us-east-1&quot;<br>}</pre><p>You can SEE what exists just by reading the code!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/0*VgwAlCfvv1uHRbaO.gif" /></figure><h3>KEY CONCEPTS OF IaC</h3><h4>Concept 1: Resources</h4><p>A <strong>resource </strong>is anything in the cloud.</p><p><strong>Example</strong>: Storage bucket, Database, Virtual Machine and so on!</p><h4>Concept 2: State</h4><p><strong>THE MOST IMPORTANT CONCEPT IN IaC!</strong></p><p>A state is a memory of what exists in the cloud.</p><p><strong>Why State Is Important:</strong></p><pre>Without state:<br><br>You: &quot;Create a bucket&quot;<br>Terraform: *Creates bucket*<br><br>You: &quot;Create a bucket&quot; (again)<br>Terraform: &quot;Okay!&quot; *Creates ANOTHER bucket* (bad!)<br><br><br>With state:<br><br>You: &quot;Create a bucket&quot;<br>Terraform: *Creates bucket, saves to state*<br><br>You: &quot;Create a bucket&quot; (again)<br>Terraform: &quot;Already exists in state. No changes needed.&quot;</pre><p>You can store states locally (on your computer) or remotely (in cloud storage; better for teams, better for production).</p><h4>Concept 3: Providers</h4><p>Providers are the plugins that talk to the cloud. Service Providers literally!</p><ul><li>Google Cloud Provider = The driver for Google Cloud</li><li>AWS Provider = Driver for AWS</li><li>Azure Provider = Driver for Azure</li></ul><h4>Concept 4: Modules</h4><p>Modules are reusable chunks of infrastructure code. Just like functions in programming!</p><blockquote>DRY = Don’t Repeat Yourself!</blockquote><h4>Concept 5: Dependencies</h4><p>The order is figured automatically!</p><pre>resource &quot;google_storage_bucket&quot; &quot;data&quot; {<br>  name = &quot;my-bucket&quot;<br>}<br><br>resource &quot;google_bigquery_dataset&quot; &quot;analytics&quot; {<br>  dataset_id = &quot;analytics&quot;<br>  <br>  # This references the bucket<br>  default_table_expiration_ms = 3600000<br>  <br>  # Terraform knows to create bucket FIRST, then dataset<br>  # ...except explicitly told to do otherwise</pre><h3>What does an IaC Workflow look like?</h3><p>Using Terraform as an example, a typical workflow has 4 steps:</p><pre>1. Initialize - `terraform init`<br># Downloads providers, sets up workspace. You run this FIRST.<br><br>2. Plan - `terraform plan`<br># Shows what will happen WITHOUT making changes. Like a preview!<br><br>3. Apply - `terraform apply`<br># Actually creates/modifies/deletes infrastructure. <br># You type &quot;yes&quot; to confirm.<br><br>4. Destroy - `terraform destroy`<br># Tears everything down. Gotta use carefully!</pre><p>Think of it like:</p><p>init = Installing dependencies</p><p>plan = Reviewing your shopping cart</p><p>apply = Your Place Order button</p><p>destroy = Returning everything</p><p><strong>It typically looks like this:</strong></p><pre>Let&#39;s create a storage bucket in Google Cloud:<br><br>**main.tf:**<br>```<br>terraform {<br>  required_providers {<br>    google = {<br>      source = &quot;hashicorp/google&quot;<br>    }<br>  }<br>}<br><br>provider &quot;google&quot; {<br>  project = &quot;learning_IaC&quot;<br>  region  = &quot;northamerica-northeast1&quot;<br>}<br><br>resource &quot;google_storage_bucket&quot; &quot;data_lake&quot; {<br>  name     = &quot;my-data-lake&quot;<br>  location = &quot;canada&quot;<br>}<br>```<br><br>**Then run it:**<br>```bash<br>terraform init    # Setup<br>terraform plan    # Preview<br>terraform apply   # Create!<br>```<br><br>**Result:** You have a cloud storage bucket!</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*vBMfTgtiL7bILAzP.gif" /></figure><h3>Why should you care as a data engineer/ML person?</h3><p>Remember my Docker article where we containerized applications? IaC is the next level!</p><p><strong>Docker</strong> = Packages your code</p><p><strong>IaC</strong> = Creates the infrastructure to run that code</p><p>For data pipelines:</p><p>— Terraform creates your cloud storage buckets</p><p>— Terraform provisions your databases</p><p>— Terraform sets up compute resources</p><p>— Docker packages your pipeline code</p><p>— Everything runs together!</p><p>— Everybody is happy!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/320/0*23_NmxXEQf9jpHtH.gif" /></figure><p>When you move from local Jupyter notebooks to production cloud pipelines, IaC is what makes it possible.</p><h3>IN SUMMARY:</h3><h4>Infrastructure as Code Means:</h4><ol><li><strong>Writing code</strong> to define infrastructure</li><li><strong>Version control</strong> infrastructure like software</li><li><strong>Review changes</strong> before applying</li><li><strong>Automate</strong> infrastructure creation</li><li><strong>Reproduce</strong> environments consistently</li><li><strong>Collaborate</strong> with teams effectively</li></ol><p>Thank you for coming to my TEDTALK!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=41c89a72b8d9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is Docker and Why Do We Use It?]]></title>
            <link>https://zainabmosunmola.medium.com/what-is-docker-and-why-do-we-use-it-2d663cde844f?source=rss-334bc9672bd4------2</link>
            <guid isPermaLink="false">https://medium.com/p/2d663cde844f</guid>
            <category><![CDATA[data-engineering]]></category>
            <category><![CDATA[backend]]></category>
            <category><![CDATA[software-architecture]]></category>
            <category><![CDATA[docker]]></category>
            <category><![CDATA[software-engineering]]></category>
            <dc:creator><![CDATA[Zainab Mosunmola]]></dc:creator>
            <pubDate>Sun, 18 Jan 2026 00:56:49 GMT</pubDate>
            <atom:updated>2026-01-27T10:27:34.913Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/736/0*yP89Bs2aSWFSd0ZI.jpg" /><figcaption>Source: Pinterest</figcaption></figure><p>I started exploring Docker, and almost every definition I found left me confused. As a girlie whose entry into tech was through Data Science, I was familiar with words like <em>server</em>, <em>VM</em>, and <em>backend,</em> but I didn’t really understand how they all connected.</p><p>So, imagine my confusion when I found the definition:</p><blockquote>Docker is a <em>containerization software</em> that allows us to isolate software in a similar way to virtual machines but in a much leaner way.</blockquote><p>In simpler terms:</p><blockquote>Docker lets you box your software so it runs anywhere, like a virtual machine, but lighter and faster. It puts everything needed to run your application together in this box, where your app does not interfere with other apps and other apps do not interfere with yours, either. So anywhere the application runs, it behaves the same way and does not break.</blockquote><p>…or at least that was what I got from all ChatGPT’s rambling.</p><p>Let me break it down the way I wish I had it broken down for me.</p><h4>Servers and the Cloud</h4><p>At the bottom of everything, we have <strong>servers</strong>.<br>Real metal computers that are owned by companies like Amazon, Google, and Microsoft.</p><p>Grouped? You have data centers. When those data centers are exposed to you over the internet, we call that <strong>the cloud</strong>.</p><p>The cloud is basically:</p><blockquote>A company that owns many servers and rents them to you online.</blockquote><p>This is where we have Amazon Web Services (AWS), Google Cloud, and Microsoft Azure.</p><p>When you use the cloud, we are renting:</p><ul><li>servers</li><li>storage</li><li>GPUs</li><li>databases</li><li>networking, and so on…</li></ul><p>…which are all built into the real physical machines.</p><h4>Where Virtual Machines Come In</h4><p>Now, what you rented is presented to you as a Virtual Machine (VM).</p><p>A VM is a virtual computer carved out of the physical server.<br>Inside that VM, an <strong>operating system (OS)</strong> runs.</p><p>The operating system manages:</p><ul><li>processes</li><li>memory</li><li>networking</li><li>disk access.</li></ul><p>This is where your backend runs.</p><h4>Frontend vs Backend</h4><p>Let’s step aside for a bit.</p><p>When you build an application, you usually have two parts:</p><ul><li><strong>Frontend</strong><br>This is what you download on your phone or open in your browser. This is what you interact with directly.</li><li><strong>Backend</strong><br>This handles logic, authentication, databases, payments, and computation.</li></ul><p>The frontend communicates with the backend over the internet.</p><p>The backend runs on servers.</p><p>Specifically, the backend runs as <strong>processes on the server’s operating system</strong>.</p><h4>Where Docker Comes In</h4><p>This backend process can be packaged into a <strong>container</strong> with everything it needs to stay jiggy.</p><p>Here, my friends, is where Docker comes in.</p><p>Docker is software that helps you package your application, its libraries, its runtime, and its environment into a container so it runs the same way on any server operating system.</p><p><strong>Note </strong>that Docker is just one option for running backends, and it is not required.</p><p>Backends can run:</p><ul><li>Directly on servers (no containers)</li><li>In containers (Docker)</li><li>In VMs</li><li>or Serverless</li></ul><p>Now, let’s step back in front 😂 .</p><h4>Why Is Docker Compared to Virtual Machines?</h4><p>This part confused me because it sounded like Docker was an improved version of VMs, like a parallel comparison.</p><p>The keyword is <strong>isolation</strong>.</p><p>Before Docker existed, virtual machines were used to isolate software. A VM isolates:</p><ul><li>virtual CPU</li><li>virtual memory</li><li>virtual disk</li><li>and an operating system</li></ul><p>When Docker was created, isolation was also its theme, but it had a different goal.</p><p>Docker isolates software <strong>in a container</strong>.</p><p>A container houses:</p><ul><li>pipelines</li><li>processes</li><li>file systems</li><li>libraries</li><li>network space</li><li>runtime.</li></ul><p>Unlike VMs, there is no virtual hardware, no separate OS. The same model fairly.</p><h4>In conclusion</h4><p>Docker is not the same as a virtual machine nor parallel. It is a way to package software processes together with their dependencies so they can run consistently on any server operating system.</p><p>Virtual machines isolate hardware.<br> Containers isolate processes.</p><p>That’s it.</p><p>Thank you for coming to my TED TALK!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2d663cde844f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Shipping Data Centers to Space]]></title>
            <link>https://zainabmosunmola.medium.com/shipping-data-centers-to-space-872fcd8348d0?source=rss-334bc9672bd4------2</link>
            <guid isPermaLink="false">https://medium.com/p/872fcd8348d0</guid>
            <category><![CDATA[space]]></category>
            <category><![CDATA[data-center]]></category>
            <category><![CDATA[engineering]]></category>
            <category><![CDATA[physics]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Zainab Mosunmola]]></dc:creator>
            <pubDate>Wed, 19 Nov 2025 05:18:31 GMT</pubDate>
            <atom:updated>2025-11-19T05:18:31.234Z</atom:updated>
            <content:encoded><![CDATA[<h4>A clear technical breakdown of the 2024 white paper “Why We Should Train AI in Space by Starcloud.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5bJ5NlRVp18YCF6H9tSMug.png" /></figure><p>If you have been on the internet long enough, you will know that the next scary thing after “AI will take over our jobs” is “AI will destroy the planet,” and this particular one is a real concern, I can’t lie.</p><p>At this point, not just the masses are concerned; the whole tech ecosystem is worried, including those at the top.</p><p>So what is the plan?</p><p>This weekend, I reviewed Starcloud’s white paper and their ambitious plan to ship data centers to space.</p><p>And let me just say: the audacity is crazy… but the math is mathing.</p><h3>1. Why This White Paper Exists</h3><p>Artificial intelligence is scaling at a very fast pace, faster than we can keep up with. Every increase in capability leads to exponentially larger compute clusters, more GPUs, more networking bandwidth, and far more electricity.</p><p>This has created a new reality:</p><blockquote><strong><em>The progress of AI development is now directly constrained by global energy availability.</em></strong></blockquote><p>Some of the biggest names in technology, Sam Altman, Elon Musk, and Mark Zuckerberg, have all flagged the same issue: <strong>we are running out of electricity for AI</strong>.</p><p>We can’t build new capacity fast enough, and traditional data centers are hitting physical limits in cooling, land, water, and power.</p><p>Starcloud’s white paper proposes an unconventional solution:</p><p><strong>Ship the whole damn thing to Space!</strong></p><p>Move the largest AI training clusters into orbit, where sunlight and cold are abundant.</p><p>They are not just proposing training AI in space; they argue that doing so is necessary because <strong>Earth will not be able to deliver the energy scale AI needs.</strong></p><p>Here, I will unpack what they claim, how it works, and the challenges that might be hidden between the lines.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*aDUhxvQMilmuSjfe.jpg" /><figcaption>Source: Tom’s Hardware</figcaption></figure><h3>2. The Problem: AI’s Appetite is Getting Out of Hand</h3><h3>2.1 AI scaling laws changed the game</h3><p>AI performance scales with compute.</p><p>More data + more parameters + more compute = better performance.</p><p><strong>Let me give you an analogy:</strong></p><ul><li>A 2020 frontier model may use <strong>10–20 MW</strong> of power for training.</li><li>A 2024 frontier model approaches <strong>50–100 MW</strong>.</li><li>A 2026–2027 frontier model is expected to require <strong>1–5 GW</strong> for training alone.</li></ul><p>The paper’s estimate for <strong>5 GW for GPT-6/Llama 5 era models</strong> is not even based on vibes.<br> It comes from:</p><ul><li>the trend of doubling compute every 3–6 months,</li><li>training on multimodal, longer-context datasets,</li><li>the shift to agentic systems requiring persistent training cycles.</li></ul><p>Let’s not even get started with models that will come after these.</p><p>A 5 GW data center is not “big”, it is huge!</p><p><strong>It is larger than a nuclear plant</strong>, larger than most hydroelectric facilities, and far beyond the power capabilities of any commercial data center on Earth today.</p><h3>2.2 Earth can’t keep up</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*K-6ZwzQ9ZMWK0-kG.jpg" /><figcaption>Google Data Center | Source: brightlio</figcaption></figure><ul><li>Grid upgrades take <em>years</em> of planning, getting permits, and construction.</li><li>Even hyperscale data centers (Amazon, Meta, Google) face multi-year delays for power.</li><li>Local communities push back due to water use, noise, and heat emission.</li><li>Land availability and zoning approvals are additional problems.</li></ul><p>And AI is not the only thing stressing the grid:</p><ul><li>EV adoption</li><li>Electrified home heating</li><li>Green hydrogen</li><li>Industrial loads</li></ul><p>All of these push demand up.</p><p>This is why Starcloud argues that <strong>Earth cannot scale data centers at the speed AI requires.</strong></p><h3>3. The Proposal: A Gigawatt-Scale Data Center in Space</h3><p>Starcloud proposes the first <strong>multi-gigawatt AI training cluster in space</strong>, powered entirely by solar energy from orbit.</p><h3>3.1 Why Sun-Synchronous Orbit (SSO)?</h3><p>Starcloud selects a <strong>dawn–dusk SSO</strong>, which provides:</p><ul><li>Near <strong>continuous sunlight</strong></li></ul><blockquote><strong>95% capacity factor</strong></blockquote><ul><li>No atmospheric losses</li><li>No weather, clouds, or seasons</li><li>Only about 0.2% variation in irradiance all year round</li></ul><p>There won’t be a need for massive battery storage.</p><p>On Earth, solar installations typically have an average capacity factor of <strong>10–24%</strong>.</p><p>In space, it jumps to <strong>95% or higher</strong>.</p><p>Is that not even insane?!</p><p>This is a <strong>5x increase in energy output for the same panel area</strong>.</p><h3>3.2 The 4 km × 4 km thin-film solar array</h3><p>How do we get this 5 GW cluster?</p><ul><li>Starcloud assumes large-area thin-film solar panels (&lt;25 μm),</li><li>Power density &gt;1000 W/kg,</li><li>Ability to <strong>self-heal radiation damage (annealing)</strong></li><li>Fill factor 90%,</li><li>Beginning-of-life efficiency of about 22%.</li></ul><p>This results in a massive <strong>16 km²</strong> structure, let’s say a large airport.</p><p>No human-made array of this size has ever been deployed.</p><p>But in orbit, the absence of gravity allows for structures that would have collapsed on Earth.</p><p>This is not based on vibes, once again; the physics checks out.</p><h3>3.3 Space as the perfect cooling environment</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eShURQDw7yIaj0xh8nfMMQ.png" /><figcaption>Chiller</figcaption></figure><p>On Earth:</p><ul><li>Cooling is the second-largest cost in data centers</li><li>Chillers struggle past <strong>45°C</strong> days</li><li>Millions of tons of water per year</li></ul><p>In space:</p><ul><li>Background temperature = <strong>2.7 K (about −270°C)</strong></li><li>Radiators at <strong>20°C</strong> can dump heat with no mechanical energy</li><li>Radiative cooling becomes the default instead of the bottleneck</li><li>No chillers</li><li>No evaporative towers</li><li>No water (a single 40 MW data center uses 1.7 million tons of water over 10 years)</li></ul><p>This represents a fundamental shift because cooling no longer becomes a major problem.</p><h3>4. Let’s go deeper into the engineering.</h3><h4>How does the Orbital Data Center work?</h4><h3>4.1 Modular compute nodes</h3><p>Each launch sends about a 100-ton compute container with:</p><ul><li>about 300 racks</li><li>120 kW per rack</li><li>about 40 MW compute per module</li></ul><p>Training requires extremely low latency, meaning tens of nanoseconds between GPUs.</p><p>This means all containers must be physically within <strong>hundreds of meters</strong> of each other.</p><p>The cluster forms a <strong>3D lattice</strong>, not a 2D floor like on Earth.</p><p>Vacuum interconnects benefit from:</p><ul><li>35% faster speed of light (than fiber)</li><li>Shorter distances (no long hallway cables)</li><li>Reduced electromagnetic interference</li></ul><h3>4.2 Power distribution infrastructure</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9RJTnpe1Sh2LLpEgGpAAzQ.png" /></figure><p>Starcloud uses High-Voltage Direct Current (<strong>HVDC)</strong> because it’s lighter, more efficient, and directly compatible with solar power. In space, you want as few conversions and heavy components as possible, so HVDC keeps losses low and cables thin.</p><p>Power flows look like:</p><p><strong>Solar array → HVDC backbone → DC converters → racks</strong></p><p>The only real “delicate” part is the <strong>universal port</strong> that connects each compute container. It must deliver power, cooling, and high-speed networking through a single interface that operates reliably in a vacuum and won’t fail for years.</p><p>Reliability is important here because one faulty connector can throw an entire module off balance.</p><h3>4.3 Thermal management</h3><p>Thermal management has two layers in Starcloud’s architecture: <strong>internal cooling</strong> (inside each compute module) and <strong>external cooling</strong> (releasing heat into space).</p><h4>Internal:</h4><ul><li>Direct-to-chip cooling</li><li>Two-phase immersion systems (boiling liquid → vapor → condensation, which is complex in zero-gravity)</li></ul><p>These aren’t a problem on Earth, but in space, zero gravity makes these things act differently:</p><ul><li>Fluids don’t settle at the bottom of a chamber</li><li>Bubbles don’t rise naturally</li><li>Vapor doesn’t separate cleanly from liquid</li><li>Pumps must be designed to handle irregular flow paths</li></ul><p>A failure in internal cooling not only reduces efficiency, but it can also quickly destroy racks.</p><h4><strong>External</strong>:</h4><p>The radiators handle the bulk of the heat load by releasing thermal energy into space. They must be:</p><ul><li>enormous</li><li>lightweight</li><li>damage-tolerant</li><li>stable for a decade</li><li>resistant to micrometeoroids</li><li>thermally reliable at 20°C</li></ul><p>These will be the <strong>largest radiators ever built.</strong></p><h3>4.4 Launch and rapid deployment</h3><p>Deployment relies on next-gen heavy-lift rockets (Starship, New Glenn, Neutron).<br> Starcloud’s timeline assumes:</p><ul><li>$5M per launch</li><li>$30/kg to SSO</li><li>up to 3 launches/day</li><li>&lt;100 launches for the full 5 GW system</li><li>2–3 months total deployment time</li></ul><p>This represents a future where launch capacity is abundant and cheap, which is currently not the world we live in today.</p><h3>5. The Economics: Why Space Might Beat Earth</h3><p>The boldest claim in the paper is that space is <em>cheaper</em> than Earth, so let’s see.</p><h3>5.1 The 40 MW, 10-year comparison</h3><p>Terrestrial 40 MW data center (10-year view):</p><ul><li>$140M electricity</li><li>$20M backup power</li><li>$7M cooling energy</li><li>$167M total costs</li><li>1.7M tons of water used</li></ul><p>Orbital 40MW equivalent:</p><ul><li>$2M solar array</li><li>$1.2M shielding</li><li>about $8.2M total cost</li></ul><p>The difference is interesting and the reason is:</p><blockquote><strong><em>Space solar power has almost no ongoing cost.</em></strong></blockquote><ul><li>Free power (sunlight)</li><li>Free cooling</li><li>No weather.</li><li>No land consumption.</li><li>No water.</li><li>No battery storage.</li><li>No grid losses.</li><li>After installation, energy approaches about $0.002/kWh.</li></ul><p>See, Earth will never see those stats!!!</p><h3>6. The Assumptions: Where the Risks Are</h3><p>To understand the proposal honestly, we need to highlight the assumptions that must hold:</p><h3>6.1 Launch costs must collapse</h3><p>The economics depend on:</p><ul><li>$30/kg, which needs Starship-level economies.</li><li>$10/kg will require <em>industrialized,</em> fully reusable rockets.</li><li>Multiple launches per day</li><li>Unprecedented reliability</li></ul><p>Today, launch costs are still higher and cadence is lower.</p><h3>6.2 GW-scale thin-film solar is still in development</h3><p>GW-scale production of 25 μm cells with self-healing properties is not yet a thing.</p><h3>6.3 Radiators of unprecedented size</h3><p>4 km arrays + multi-km radiators = materials engineering that we have not seen yet.</p><h3>6.4 High-density cooling in zero gravity</h3><p>Two-phase systems behave very differently without gravity.<br> Fluid distribution becomes a research problem.</p><h3>6.5 Unknown lifespan of commercial GPUs in space</h3><p>Radiation shielding is estimated at:</p><ul><li>1 kg per kW of compute<br> Even small radiation errors can fry expensive chips.</li></ul><p>They estimate 10 years.<br> It could be less.</p><h3>6.6 Regulatory pressure</h3><p>Because this involves:</p><ul><li>mega-structures</li><li>orbital debris</li><li>astronomy interference</li><li>Licensing for dozens of rapid launches</li><li>SSO congestion</li></ul><p>These issues become geopolitical, and regulation could significantly slow deployment.</p><h3>7. What This White Paper Is Actually Doing</h3><p>Beyond physics and engineering, the document is making a clear strategic statement.</p><h3>7.1 They are positioning space as inevitable</h3><p>The argument is:</p><ul><li>AI dominates the century</li><li>AI requires massive energy</li><li>Earth cannot deliver</li><li>Therefore, space is next</li></ul><p>It frames orbital AI compute as a logical, scalable step, not speculation.</p><h3>7.2 They are claiming first-mover advantage</h3><p>Multiple lines emphasize:</p><ul><li>“No insurmountable obstacles”</li><li>“Working from first principles”</li><li>“First company pursuing orbital GW-scale data centers”</li></ul><p>This is investor language.</p><h3>7.3 They link themselves to four megatrends</h3><p>They anchor on:</p><ol><li>Falling launch costs</li><li>AI energy demand</li><li>Larger GPU clusters need</li><li>Cheap orbital connectivity</li></ol><p>They are building a story: <strong>The time is now, and it is urgent.</strong></p><h3>Is Training AI in Space Realistic?</h3><p>Starcloud’s proposal is ambitious and grounded in real physics. The opportunities are genuinely exciting, but the challenges are just as significant:</p><ul><li>Unproven solar tech</li><li>Gigantic radiators</li><li>Zero-g cooling</li><li>High launch cadence</li><li>Data logistics</li><li>Radiation tolerance</li><li>Regulatory constraints</li><li>thin-film solar breakthroughs,</li><li>mega-structure deployment,</li><li>regulation,</li><li>and the lifespan of terrestrial GPUs in orbit.</li></ul><p><strong>Everything has to align perfectly! </strong>Maybe not perfectly…</p><p>But if electricity demand continues to increase and constraints align as Starcloud expects, training AI in space may shift from futuristic to inevitable.</p><p>Feilden, E., Oltean, A., &amp; Johnston, P. (2024). <em>Why we should train AI in space</em>. <a href="https://starcloudinc.github.io/wp.pdf?fbclid=PAVERFWAODZ6NleHRuA2FlbQIxMQBzcnRjBmFwcF9pZA8xMjQwMjQ1NzQyODc0MTQAAac3tGuZdtPZH8wyBBTTaCEZ_e0YSBUEOg2DiZ_87aTH46FXSiFgRgW3NT45Hw_aem_MrwmevWDB_9eSx2Ow0desA">https://starcloudinc.github.io/wp.pdf</a></p><p>‌</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=872fcd8348d0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Solving the Plateau Problem and What It Means for You]]></title>
            <link>https://zainabmosunmola.medium.com/solving-the-plateau-problem-and-what-it-means-for-you-3523ecd83276?source=rss-334bc9672bd4------2</link>
            <guid isPermaLink="false">https://medium.com/p/3523ecd83276</guid>
            <category><![CDATA[engineering]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[science]]></category>
            <category><![CDATA[mathematics]]></category>
            <dc:creator><![CDATA[Zainab Mosunmola]]></dc:creator>
            <pubDate>Thu, 13 Nov 2025 05:20:19 GMT</pubDate>
            <atom:updated>2025-11-13T05:20:19.610Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*xmBsu3N179JCDpoV.jpg" /><figcaption>Munich Olympic Stadium</figcaption></figure><p>In the mid-19th century, a Belgian physicist named Joseph Plateau created a math problem that took a hundred years to solve, and scientists just solved the toughest part. An experiment that started with wireframes, soap and bubbles.</p><h4>Here is the concept:</h4><p>When you dip any shape into soapy water, a thin soap film forms, which is what we call bubbles. Before we blow it out and form a bubble, it usually just sits there as a thin film. This film tries to find the least amount of area possible.</p><p>So if you bend the wire in a circle, it would form a flat disk; if you twist the wire to form two circles, it takes the form of my waistline, you call it an hourglass shape, mathematicians call it a <strong>catenoid.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/311/0*J3ozwKZj7Y51b-u-" /></figure><p>Different wire frames, different film shapes…</p><p>…and they always try to take up the smallest area possible. Something mathematicians call an area-minimizing surface.</p><p>They literally just try to manage space by stretching themselves into the most efficient 2D shape possible around any boundary that is given.</p><p>It became a math question: <strong>If a shape is drawn in 3D space, can a soap film find the minimal surface for it?</strong></p><p>They said this is a math optimization problem and called it the <strong>Plateau Problem.</strong></p><p>I don’t know why they can’t just play with the bubbles like everyone else and leave it at that.</p><p>In the 1930s, two mathematicians said yes!</p><blockquote>For any closed curve (your wire frame) in three-dimensional space, you can always find a minimizing two-dimensional surface (your soap film) that has the same boundary.</blockquote><p>After that, everyone got obsessed with bubbles.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/400/0*IafvOWGUVuKja_oI.gif" /></figure><p>They realized it is applicable in so many places: geometry, topology, physics, biology, molecular design, you name it. It was basically just nature’s way of being efficient.</p><p>They find out up to dimension 7, everything is going “smooth” literally lol, but higher than that, they become weird. They might fold, pinch or intersect and this weird behavior was called <strong>singularities.</strong></p><p>…which makes everything more complicated, but they soon found out these singularities are rare and when they appear, they can disappear by just wiggling your wire right.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/693/1*r7dd4m4hkT0-AyWHC2wsCQ.png" /><figcaption>Source: Quantamag</figcaption></figure><p>But from dimension 9 upwards, they become weird and chaotic again and nobody knew why for a while.</p><p>But now they have finally cracked the code, they now understand how minimal surfaces work in high dimensions and even in singularities.</p><p>…and that is a big deal!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/0*e2DW-W_Ui_cj0dMD.gif" /></figure><h4>Let me tell you what that means for you:</h4><p>This affects any aspect that involves optimization, shapes, surfaces, energy, or stability.</p><ul><li>Better understanding, modeling and simulation of nature.</li><li>Better modeling of complex systems.</li><li>Improving robotics</li><li>Understand the universe better (especially things that have to do with spacetime)</li><li>Making 3D modeling and animation cleaner (think computer graphics, VR, movies, games</li><li>Train AI better (think diffusion models, neural networks, optimization algorithms)</li><li>Build lightweight but stronger materials (think the aerospace and automobile industry). I can tell you that there is nothing engineers love more than minimum material for maximum strength.</li></ul><p>So, this is not just math for math people. It is a math for everyone, math that affects the everyday person. It is math creating a kit for science, engineering, and, of course, AI.</p><p>Tinaglia, G. (n.d.). <em>The theory of minimal surfaces: Plateau Problem</em>. Retrieved November 13, 2025, from <a href="https://www.homepages.ucl.ac.uk/~ucaheps/topics/plateau.pdf">https://www.homepages.ucl.ac.uk/~ucaheps/topics/plateau.pdf</a>‌</p><p><em>New Proofs Probe Soap-Film Singularities | Quanta Magazine</em>. (2025, November 12). Quanta Magazine. <a href="https://www.quantamagazine.org/new-proofs-probe-soap-film-singularities-20251112/?utm_campaign=linkinbio&amp;utm_medium=referral&amp;utm_source=later-linkinbio">https://www.quantamagazine.org/new-proofs-probe-soap-film-singularities-20251112/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3523ecd83276" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Rabbit Hole: How Facebook Location Data Model Migration Patterns]]></title>
            <link>https://zainabmosunmola.medium.com/rabbit-hole-how-facebook-maps-human-migration-e00bae14fd96?source=rss-334bc9672bd4------2</link>
            <guid isPermaLink="false">https://medium.com/p/e00bae14fd96</guid>
            <category><![CDATA[facebook]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[rabbit-hole]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[migration]]></category>
            <dc:creator><![CDATA[Zainab Mosunmola]]></dc:creator>
            <pubDate>Wed, 01 Oct 2025 08:47:53 GMT</pubDate>
            <atom:updated>2025-10-01T11:12:57.257Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/949/1*OeDRLnjOYB3z6k4VhMv9rg.png" /></figure><p>I was reading <em>Data Science from Scratch</em> by Joel Grus when I hit this line:</p><p><em>“Facebook asks you to list your hometown and your current location, ostensibly to make it easier for your friends to find and connect with you. But it also analyzes these locations to identify global migration patterns and where the fanbases of different football teams live.”</em></p><p><strong>How?</strong></p><p>How does Facebook turn “hometown: Lagos” and “current city: London” into migration data? How do they map population movements at scale? And more importantly, what are they actually <em>doing</em> with this data?</p><p>So I went down the rabbit hole. And what I found was way more fascinating (and useful) than I expected.</p><h3>How Facebook Collects Location Data</h3><p>Let’s start with the basics. Facebook isn’t just looking at what you type in your profile. They’re pulling location data from multiple sources:</p><ul><li><strong>GPS coordinates</strong> from your phone</li><li><strong>IP addresses</strong> when you log in</li><li><strong>Wi-Fi signals</strong> and <strong>mobile cell towers</strong> nearby</li><li><strong>Geotags</strong> embedded in photos you upload</li><li><strong>Login patterns</strong> over time</li></ul><p>That last one is key. If you used to log in from Lagos every day for two years, then suddenly all your logins are from London, Facebook notices. Multiply that across thousands or millions of users, and you start to see patterns.</p><p>When enough people make similar moves, Facebook can infer <strong>population-level migration flows</strong>. Not just “this person moved,” but “this many people moved from City A to City B during this time period.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*05jKnkIr723rHilqo56KmA.png" /></figure><h3>From Data Collection to Migration Mapping</h3><p>Facebook compares two things:</p><ol><li><strong>Your stated locations:</strong> “Hometown” vs “Current Location” in your profile</li><li><strong>Your behavioral patterns:</strong> Where you logged in from months ago vs where you log in from now</li></ol><p>If these align across thousands of users, they can map out:</p><ul><li>Which cities are people leaving</li><li>Which cities are people moving to</li><li>How quickly these shifts happen</li></ul><p>This isn’t just for internal analytics or targeted ads (though it’s definitely used for that). Facebook has also started using this data for <strong>humanitarian purposes</strong> through its <em>Data for Good</em> initiative.</p><p>Which brings me to the paper that really pulled me further down this hole.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*31_txAmppRGSP6qF.gif" /></figure><h3>Case Study: Measuring Displacement After Disasters</h3><p>I found a research paper called <em>“Measuring Long-Term Displacement using Facebook Data”</em> that breaks down how Facebook tracks population displacement after natural disasters.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/634/1*65EweMvtHMYNUJy8bE8hjg.png" /></figure><h3>The Problem</h3><p>When a disaster hits (flood, cyclone, typhoon), humanitarian organizations need to answer basic questions fast:</p><ul><li>How many people were displaced?</li><li>Where did they go?</li><li>When did they return?</li></ul><p>Traditional methods struggle here. Surveys take time. Shelter counts only track people in official shelters, missing those who stay with family or friends. Census data is outdated. By the time accurate numbers come in, the crisis has often moved to a new phase.</p><p>Facebook offered an alternative: <strong>use anonymized location data from consenting users to track displacement in near real-time.</strong></p><h3>How Does It Work?</h3><p>The methodology breaks down into three phases:</p><p><strong>Before the disaster:</strong> Facebook establishes your “home” location based on where you’ve been consistently over the past 30 days. This is your baseline.</p><p><strong>After the disaster (first two weeks):</strong> If your post-disaster location is more than 2 km from your home <em>and</em> you’re moving around more than usual, you’re flagged as displaced.</p><p><strong>Long term:</strong> Facebook continues monitoring daily. If you’re observed near your home location for three consecutive days, you’re considered returned.</p><p>The data is aggregated into <strong>city-to-city flows</strong>, creating a transition matrix that shows:</p><ul><li>Where people left from</li><li>Where they went</li><li>How many returned</li><li>How many remained displaced</li><li>How many are unknown</li></ul><p>This helps responders answer the critical question: <strong>“From where → to where → for how long?”</strong></p><h3>Real Examples</h3><h3>🌪 Cyclone Fani (India &amp; Bangladesh, 2019)</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JmOSt8gMQ4-nK6Fr" /><figcaption>Source: The Guardian</figcaption></figure><p>On May 12, 2019, over <strong>94,000 Facebook users</strong> were flagged as displaced when Cyclone Fani hit.</p><p>A month later:</p><ul><li>~72,000 had returned home</li><li>~24,000 were still displaced</li></ul><p>The maps showed not just the path of the cyclone, but the <em>severity</em> of displacement in different cities. The areas closest to the cyclone’s path had the worst displacement, but the data also revealed secondary displacement patterns that official records missed.</p><h3>🌊 Typhoon Hagibis (Japan, 2019)</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/445/0*QK6aP8ssXwzHCbWF" /><figcaption>Source: The Guardian</figcaption></figure><p>When Typhoon Hagibis hit Japan, Facebook tracked displacement out of Tokyo and surrounding regions. Here’s what made this interesting:</p><p><strong>Facebook’s numbers showed more displaced people than Japan’s official shelter records.</strong></p><p>Why? Because many displaced people didn’t go to shelters. Instead, they stayed with family, friends, or in hotels. Official data typically only count people in registered shelters, which underestimates actual displacement.</p><p>Facebook also ran a survey of over 20,000 users in Japan to understand:</p><ul><li>Why people left (government order vs personal choice)</li><li>How long did they stay away</li><li>And where they went</li></ul><p>This approach, using location data and a survey, gave a broader view than using either source alone.</p><h3>What Makes This Approach Powerful</h3><p><strong>Speed:</strong> They got daily updates on displacement patterns, not weeks or months.</p><p><strong>Coverage:</strong> They were able to detect displacement outside of the official shelters, which helps give a more realistic perspective of how disasters affect populations.</p><p><strong>Longitudinal tracking:</strong> They were able to monitor long-term recovery, not just immediate evacuation. You can see when people return, when new displacement happens, and when recovery is stalling.</p><p><strong>Data gaps:</strong> Fills in blanks where official numbers don’t exist or are incomplete.</p><p><strong>Complementary:</strong> It is important to know it works alongside traditional methods like surveys and census data, and not to serve as a replacement.</p><h3>Privacy Protections</h3><p>Given how sensitive this data is, Facebook implemented several safeguards:</p><ul><li>Data is <strong>anonymized and aggregated</strong></li><li>Only <strong>city-level summaries</strong> are shared (no individual tracking)</li><li>To protect identity, transitions with less than 10 people are <strong>hidden</strong></li><li><strong>No cross-border movement</strong> was reported to avoid political problems</li><li>It only included users who <strong>turned on </strong>their location</li></ul><h3>The Limitations</h3><p>This approach is still not perfect, there are still some important drawbacks:</p><p><strong>Sampling bias:</strong> Only includes Facebook users who enabled their location. This skews toward younger and more tech-savvy people with access the internet.</p><p><strong>Coverage gaps:</strong> Doesn’t capture everyone displaced, especially vulnerable populations without internet access.</p><p><strong>Distance threshold:</strong> Doesn’t detect people who move less than 2 km from home, even if they’re displaced.</p><p><strong>Infrastructure dependency:</strong> Requires working internet and cell towers, which may be damaged in disasters.</p><p><strong>Platform bias:</strong> People who don’t use Facebook (or have deleted it) aren’t represented.</p><p>Despite these limitations, the approach provides a valuable signal where traditional data sources fall short.</p><h3>Why This Matters</h3><p>Facebook figured out how to track, summarize, and visualize population movement during and after disasters using only anonymous data from consenting users. Relief agencies can now know:</p><ul><li>How severe displacement is</li><li>Where help is most needed</li><li>How long it takes for communities to recover</li></ul><p>This is data science in action. Not just building recommendation systems or optimizing ad revenue, but helping organizations save lives and allocate resources more effectively.</p><p>The same principles apply to other migration patterns, too, like economic migration, climate displacement, urbanization trends. The methodology Facebook developed for disaster response has broader implications for understanding how and why populations move.</p><h3>The Bigger Picture</h3><p>We generate location data constantly. Every time we use our phones, connect to Wi-Fi, post a photo, check in somewhere. That data exists whether we think about it or not.</p><p>The question isn’t whether this data will be collected anymore. It’s how it will be used.</p><p>Facebook’s disaster displacement work shows one side of that equation: data used to understand crises and coordinate responses. But the same infrastructure that enables this also enables surveillance, targeted manipulation, and privacy erosion.</p><p>Understanding how these systems work, what they’re capable of, and what their limitations are matters. Not just for data scientists, but for anyone living in a world where our movements leave digital traces.</p><p>That’s why I went down this rabbit hole. And I’m glad I did.</p><p><strong>Further Reading:</strong></p><ul><li><em>Data Science from Scratch</em> by Joel Grus</li><li><a href="https://www.internal-displacement.org/global-report/grid2020/downloads/background_papers/2020-IDMC-GRID-background-facebook-data.pdf?utm_source=chatgpt.com">“Measuring Long-Term Displacement using Facebook Data” (Facebook Data for Good)</a></li><li>Facebook Data for Good: <a href="https://dataforgood.facebook.com/">https://dataforgood.facebook.com/</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e00bae14fd96" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[USING AI TO IMPROVE AIRCRAFT DE-ICING EFFICIENCY AND SAFETY]]></title>
            <link>https://zainabmosunmola.medium.com/using-ai-to-improve-aircraft-de-icing-efficiency-and-safety-5017bd9a3b7f?source=rss-334bc9672bd4------2</link>
            <guid isPermaLink="false">https://medium.com/p/5017bd9a3b7f</guid>
            <category><![CDATA[disney]]></category>
            <category><![CDATA[aerospace]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[aircraft]]></category>
            <dc:creator><![CDATA[Zainab Mosunmola]]></dc:creator>
            <pubDate>Wed, 13 Aug 2025 09:52:35 GMT</pubDate>
            <atom:updated>2025-08-13T09:52:35.169Z</atom:updated>
            <content:encoded><![CDATA[<blockquote><em>The girlies said catch flight, not feelings, and I listened. Unfortunately, my flight was out here catching ice like Elsa. So today we’re talking Intelligent Aircraft De-Icing using AI because planes shouldn’t be out here acting like Disney princesses.</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*B7szLhuLkc6u3Nk_.jpg" /></figure><p>See, I took the plane for the first time, and to say it was a great experience is an understatement. I observed everything, every little detail. I was disappointed that I didn&#39;t get the window seat because I just wanted to soak in everything, both inside the plane and outside. I finally got my chance during the flight from Addis Ababa to Toronto; the plane wasn’t full, and I finally got the chance to switch seats!</p><p>One of the FASCINATING THINGS I OBSERVED was ice crystals on the window pane. I was just fascinated; I didn’t think too much of it till I was giving a breakdown of my experience to my gist buddy, CHATGPT🌚 (I know what you are thinking, but that is not why we are here😌).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/585/1*tP2_oKHqqahhHli_BOcurA@2x.jpeg" /><figcaption><a href="https://1drv.ms/v/c/1c42e50dde601d71/EeSFe6TytiBDiDRpkguggSEBBbdK5RMB2TI9pMVOiXHr8w?e=t0bbCV">see full video here</a></figcaption></figure><p>It was like “Did you know there’s research on using AI to predict ice formation on airplane wings in real time?” What??? And you know me, Aerospace? AI? Research? My doggy research ears are fully erect! Put me on, please!!</p><p>Basically, the plane can “think” where ice will form and adjust on the fly. Wild! I found two research papers on it, and I’ll break them down later. But first, let’s gist about why this even matters.</p><blockquote><a href="https://www.mdpi.com/2076-3417/15/14/7752">Real-Time Image Analysis for Intelligent Aircraft De-Icing Decision Support Systems by Sylwester Korga</a></blockquote><blockquote><a href="https://www.mdpi.com/2226-4310/10/8/676">Machine Learning and IoT Trends for Intelligent Prediction of Aircraft Wing Anti-Icing System Temperature</a></blockquote><h4>WHY DOES IT MATTER?</h4><p>So as the plane moves, tiny ice crystals in the air stick to the leading edge of the wings. This can literally form within minutes to hours, depending on the temperature, humidity, and altitude. If this isn’t detected and removed, it can accumulate on the edges and control surfaces, which changes the aerodynamics and reduces efficiency. This can, in turn, affect the plane’s lift, drag, and overall performance, especially during takeoff and landing.</p><p>So to solve this, the researchers are using AI to predict and manage the ice formation on wings. It does this through real-time ice detection, predictive modelling, and smart anti-icing systems. In layman&#39;s terms, they just combine AI with traditional anti-icing methods.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eZ1GQUHuMwSxDKRh3v7Kzg.png" /></figure><h4>HOW MUCH ICE IS ACTUALLY DANGEROUS?</h4><p>As little as a thin layer of ice, a sheet of frost on a windshield, can mess with a plane’s aerodynamics. I mean, if you are familiar with the aerospace industry, you’d understand how important aerodynamics is. This reduces the lift (this is the force that keeps the plane in the air), increases the drag (this makes the plane work harder to move forward), and can cause control issues if it builds on ailerons, elevators, or rudders which are primary flight control surfaces on a plane and are essential for safe and coordinated flight.</p><h4>HOW DO PILOTS HANDLE THIS MID-FLIGHT?</h4><ol><li><strong>De-Icing Systems</strong>: Most planes have heated surfaces or inflatable boots on the wings that break ice off.</li><li><strong>Anti-Icing Fluids</strong>: Some planes pump warm fluids through the leading edges to prevent ice build-up.</li><li><strong>Flight Path Adjustments</strong>: Pilots can climb or descend to avoid the worst icing clouds.</li></ol><h4>So this is where AI comes in…</h4><p>The AI systems will <strong>predict when and where ice will form</strong> so pilots can act before it becomes a problem. Some can even be programmed to <strong>control the anti-icing systems automatically</strong> in real-time. So instead of reacting after ice is visible, the plane is <em>thinking ahead</em>, like:</p><blockquote><em>“There’s a 70% chance ice will form on the left wing in 10 minutes — start heating now.”</em></blockquote><p>This is AI giving the plane a sixth sense. Ice can form faster than a pilot can say Jack Robinson, let alone react, which is why prediction tools are everything.</p><p>— — —</p><p>This was just the tip of the wing (pun intended😂😌🤭). I’ll be breaking down the two research papers in a follow-up post because trust me, they deserve their own spotlight. I will also explain the sensors and the AI algorithms behind this.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5017bd9a3b7f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How CERN’s High-Efficiency Klystron (HEK) project Can Make Netflix Better]]></title>
            <link>https://zainabmosunmola.medium.com/how-cerns-high-efficiency-klystron-hek-project-can-make-netflix-better-feb520a9321b?source=rss-334bc9672bd4------2</link>
            <guid isPermaLink="false">https://medium.com/p/feb520a9321b</guid>
            <category><![CDATA[netflix]]></category>
            <category><![CDATA[physics]]></category>
            <category><![CDATA[movies]]></category>
            <category><![CDATA[internet]]></category>
            <category><![CDATA[betrayal]]></category>
            <dc:creator><![CDATA[Zainab Mosunmola]]></dc:creator>
            <pubDate>Tue, 22 Jul 2025 00:41:31 GMT</pubDate>
            <atom:updated>2025-07-22T00:41:31.355Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>Because someone has to make sure your Wi-Fi doesn’t betray you like Efe did Oboz. (Instant classic, I can’t lie 😭)</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/370/1*fRu8p5pMiCNZkN1rJoe-FQ@2x.jpeg" /></figure><p>At first glance, particle physics and Netflix don’t exactly seem like besties. CERN is busy smashing particles and chasing the mysteries of the universe. You’re just trying to stream “To Kill A Monkey” in peace.</p><p>There’s actually a connection and it starts with how much energy it takes to run particle colliders.</p><p>It’s not direct, but here’s the logic trail:</p><p>1.	CERN is building energy-efficient tech to power large-scale computing and data systems — like those used in particle physics, cloud infrastructure, and (surprise) streaming services.</p><p>2.	As this radiofrequency tech improves, it spills over into other fields:</p><p>•	Telecom (5G, satellite internet)</p><p>•	Data transmission (faster broadband, lower latency)</p><p>•	Cloud infrastructure &amp; edge computing</p><p>3.	Netflix, YouTube, Spotify — all rely on global data networks and giant energy-hungry servers to deliver content to your screen.</p><p>4.	When those underlying systems — RF amplifiers, transmitters, even cooling systems — get more efficient, it means:</p><p>•	Lower data delivery costs</p><p>•	Greener infrastructure</p><p>•	Smoother streams without frying the planet.</p><p>—</p><h4>TL;DR:</h4><p>Better klystrons ➝ Better infrastructure ➝ Cheaper Netflix &amp; chill ➝ Greener Earth ➝ Everyone is happy.🙂‍↕️😁</p><p>So no, CERN isn’t working with Kemi Adetiba, but they might just be helping you stream her masterpiece faster, cheaper, and with a smaller carbon footprint.</p><p>You’re welcome. 😌📡🍿</p><blockquote><em>Now, can we talk about the scene in “To Kill A Monkey” where Efe betrayed Oboz. The scene broke me😭</em></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=feb520a9321b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How CERN Is Making Particle Physics Greener.]]></title>
            <link>https://zainabmosunmola.medium.com/how-cern-is-making-particle-physics-greener-654c02b984e5?source=rss-334bc9672bd4------2</link>
            <guid isPermaLink="false">https://medium.com/p/654c02b984e5</guid>
            <category><![CDATA[5g]]></category>
            <category><![CDATA[science]]></category>
            <category><![CDATA[edge-computing]]></category>
            <category><![CDATA[netflix]]></category>
            <category><![CDATA[physics]]></category>
            <dc:creator><![CDATA[Zainab Mosunmola]]></dc:creator>
            <pubDate>Mon, 21 Jul 2025 23:51:35 GMT</pubDate>
            <atom:updated>2025-07-21T23:51:35.624Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>Big Science, Big Energy Bills and the Microwave Gun That’s Getting a Makeover.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/709/1*k_jKqeIXQvylBgwyRaNKAg@2x.jpeg" /><figcaption>Credit: CERN</figcaption></figure><blockquote><a href="https://cerncourier.com/a/powering-into-the-future/?fbclid=PAQ0xDSwLrq4NleHRuA2FlbQIxMQABp1j7Y-YZv5DFY8aaUQi_G3Psr222K1MYuBnDmhrULbERzGFUBGjbIkshePuC_aem_kMrHbMckxruRocIjY_VZ1Q">Why CERN is obsessed with making your microwave gun more efficient – and why it actually matters (yes, even to you).</a></blockquote><p>Particle colliders – the massive machines that help us study things like the Higgs boson – use a lot of energy.</p><p>Like, half of that energy just goes into speeding up tiny particles using high-tech boosters called radiofrequency (RF) cavities.</p><p>These RF cavities are powered by devices called klystrons.</p><h4><strong>What’s a klystron?</strong></h4><p>Basically a radio wave amplifier from the 1930s that still runs the show in high-energy physics, broadcasting, satellites – you name it.</p><p><strong>Think</strong>: a supercharged microwave gun that turns tiny radio signals into massive electric waves to push particles to crazy speeds.</p><p>They’re old but gold.</p><p>But here’s the catch: Klystrons are energy hogs.</p><p>They usually waste ~40% of the electricity they consume. That’s like paying for a full tank and spilling nearly half before you even drive off.</p><p>Not cute. Not sustainable.</p><p>So, CERN + Lancaster Uni launched a project:</p><p>HEK – High-Efficiency Klystrons.</p><p><strong>Mission?</strong> Make klystrons that waste way less power.</p><p>They’ve already hit a 10% improvement. Now they’re gunning for 80 – 90% efficiency.</p><p>…and Yes, that’s huge!</p><p>Enter the <strong>triston</strong> – a weird, almost-forgotten cousin of the klystron pulled out of RF engineering history books.</p><p>CERN thinks it could be the key to hitting 90% efficiency – a game-changer for colliders and future tech beyond physics.</p><h4><strong>Why should you care?</strong></h4><p>Because better klystrons =</p><p>✅ Lower energy use</p><p>✅ Cheaper science</p><p>✅ Smaller environmental footprint</p><p>✅ Better tech for satellites, broadcasting, even medical machines</p><p>Your Netflix stream might just thank CERN one day.</p><h4>In short:</h4><p>CERN is revamping a 1930s technology to make the science of the future more sustainable – and the impact could reach way beyond particle colliders.</p><p>Quiet revolution. Big implications.</p><p>–</p><p>You made it to the end? Congrats, you now understand one of the geekiest energy problems in science. You’re elite. 🔋✨😌🤝</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=654c02b984e5" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>