<?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 Mike Chung on Medium]]></title>
        <description><![CDATA[Stories by Mike Chung on Medium]]></description>
        <link>https://medium.com/@miketcchung?source=rss-5fcb37186b3a------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*kyla1Cj38XKzr_XMtGPQjQ.jpeg</url>
            <title>Stories by Mike Chung on Medium</title>
            <link>https://medium.com/@miketcchung?source=rss-5fcb37186b3a------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 17:03:54 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@miketcchung/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[An Honest Review of the TryHackMe AI1 Certification]]></title>
            <link>https://meetcyber.net/an-honest-review-of-the-tryhackme-ai1-certification-e90e23bb9fed?source=rss-5fcb37186b3a------2</link>
            <guid isPermaLink="false">https://medium.com/p/e90e23bb9fed</guid>
            <category><![CDATA[agentic-ai-security]]></category>
            <category><![CDATA[cyber-security-training]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[tryhackme]]></category>
            <category><![CDATA[ai-security]]></category>
            <dc:creator><![CDATA[Mike Chung]]></dc:creator>
            <pubDate>Sun, 18 Oct 2026 09:39:31 GMT</pubDate>
            <atom:updated>2026-10-18T09:39:31.182Z</atom:updated>
            <content:encoded><![CDATA[<p>Is it the right certification for you? The good and the bad.</p><p>After a 2-day grind, I finally completed the TryHackMe AI1 certification with a passing score.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*F3G8K0IP7ezRFas8B79SNA.png" /><figcaption>Proof I passed the AI1 Certification.</figcaption></figure><p>TL;DR, here’s what I found good and bad about the certification.</p><p><strong>The Good:</strong></p><ul><li>The learning path is well-structured and informative, especially the Prompt Security and AI Supply Chain Security modules.</li><li>Easy to understand and progress through even if you do not have many years of experience or a degree in IT.</li><li>The exam is not just multiple-choice questions and fill in the blanks, there are scenarios with diagrams and interactions with AI agents.</li><li>You get your result right away after the exam. No waiting and agonizing.</li></ul><p><strong>The Bad:</strong></p><ul><li>The Threat Modeling and Data Poisoning modules are conceptually understandable but difficult to apply to the rooms and exam scenarios.</li><li>Although the exams are hands-on labs, some of the exercises are highly simulated and may not reflect what you would see in the real world.</li><li>This is a long certification. You have 48 hours to finish it and you can expect to spend 6–8 hours completing all the scenarios. Expect to write A LOT.</li></ul><p>If you want more details, read on below.</p><h3>The Good: Prompt Security and AI Supply Chain Security Modules</h3><p>First off, I need to give credit to TryHackMe’s platform<strong> </strong>and how easy it is to use. Everything is handled behind the scenes so you can put your focus only on cybersecurity tasks. As someone who is currently building their own home lab, I can testify how difficult it can be to set up a working network with multiple machines. TryHackMe handles this for you and lets you spin up machines and AI agents in just a few minutes with a click of a button.</p><p>Everything is also presented to you in bite-sized rooms that you can complete in 30–60 minutes. After learning each concept, you answer a few fill-in-the-blank questions and at the end of the room, there is a lab for you to test out your newly learned skills. This helped keep me motivated to finished the learning path.</p><p>Within the AI Security learning path, I found the Prompt Security and AI Supply Chain Security modules the most well-made. The Prompt Security labs involved doing actual prompt injection attacks on TryHackMe’s LLMs to get a flag, which was fun because it felt like I was hacking something.</p><p>The AI Supply Chain Security lab was also enjoyable and involved using scanning tools on AI models and observing the output. The process of how model vulnerabilities could be scanned was clearly described and it was straightforward to interpret the scan outputs. After these two modules, I came away with concrete techniques that I can use to manipulate LLMs for red team purposes and a set of tools I can use to assess AI model vulnerabilities before putting them in production.</p><h3><strong>The Bad: Threat Modeling and Data Poisoning Modules</strong></h3><p>The Threat Modeling and Data Poisoning modules contained useful information, but I thought the labs and scenarios in these modules were not so amazing.</p><p>For example, I was able to understand each of the OWASP Top 10 LLM risks and how to use MITRE ATLAS to understand the tactics and techniques of attackers targeting AI systems. However, the lab exercises were somewhat subjective and arbitrary, and required a specific answer to pass it. One of the exercises asks you to rate the severity of each of the vulnerabilities as either Low, Medium, or High, but nowhere in the module does it give you a systematic process to rate severity. Unless you’re using a standardized format, rating severity can be somewhat subjective so the marking system should also reflect the subjectivity and not only accept a certain answer.</p><p>Another of the exercises asks you to match which security controls would be best for which component in the AI architecture. There were some controls that I thought could be applicable to more than one component and therefore I got a lot of the answers wrong in the exercise. These components and controls are also interconnected so it is difficult to isolate its effect on just a single one.</p><p>To improve these modules, I think TryHackMe can consider either giving a concrete step-by-step procedure to assess threats and vulnerabilities, or change the marking system to allow a wider range of answers give the subjectivity.</p><p>The labs in the modules also tended to be more static sites compared to interacting with an AI system and trying to get a flag.</p><h3>The Good: You do not need much background in IT</h3><p>I felt the AI Security learning path did an excellent jobs of explaining the basics of how AI works and the theory behind how they can be exploited. Since my background is in data science and machine learning, I already had some of this knowledge so I found myself skimming through these sections. Nevertheless, for someone new to AI, it would be a high quality introduction that gives you the key points with weighing you down with the details.</p><p>Most of the concepts in the introductory AI Fundamentals module would be understandable even if you do not understand things like how TCP and UDP work, or other in-depth IT knowledge.</p><p>The more advanced modules such as the AI Supply Chain Security one would benefit from knowing things like how HTTP or API request works, but is still manageable if you don’t have expertise in these areas.</p><h3>The Good: Hands-on exam</h3><p>One of the main advantages of the AI1 certification, in my opinion, is that it requires you work through labs in a hands-on environments, as opposed to multiple-choice or fill-in-the-blank questions in traditional exams. This tests not only your knowledge of the concepts, but also if you can apply them to different scenarios, and if you can utilize technical tools and write commands.</p><p>The scenarios in the exam consisted of prompting LLMs to find exploits and obtain a flag, and also static sites with diagrams, model cards, and scenario descriptions. The ones where you have to prompt LLMs to obtain a flag is especially satisfying once you succeed in the exploit.</p><h3>The Bad: Highly simulated exercises</h3><p>I thought the exercises relied heavily on static sites with simplistic scenarios, some of which would not likely occur in the real world.</p><p>For example, when evaluating whether a model is safe to download and implement, the information shown in the simulated model cards only contains certain fields and not the amount of information that you would see on a model repository like Hugging Face.</p><p>The supply chain security scenarios also had the commands written out for you, so you didn’t have to remember any of the tools you could use to scan models. Although I do appreciate they want to make it user-friendly, I would have liked more interaction such as having to open a terminal and type the command in yourself.</p><h3>The Bad: Long exam</h3><p>This was an excruciating long exam with a lot of writing. For many of the scenarios, you are expected to justify your answer and need to write 1–2 paragraphs explaining why you selected that security control, or why you think that is the highest risk vulnerability.</p><p>You are given 48 hours to finish the exam, and the website says that you can expect to spend 6–8 hours. I usually take exams fast but I found myself spending closer to the 8-hour mark. I wouldn’t be surprised if there are people spending 9–10 hours to finish the exam, especially given the amount of writing you have to do.</p><p>You get a retry for the exam if you fail the first time, but I suggest trying to study and practice enough so you can pass on the first try. Otherwise, you might need to spend another 2 days working on the exam for your retry.</p><h3>The Good (and Bad): You get your results right away</h3><p>This is mostly good, since you don’t have to way for days or weeks to find out if you passed or not. Other professional organizations might take time to grade your exam and then send the results to you.</p><p>One drawback is that the way they get a result immediately to you is that they use AI for their grading. Especially since you have to write a lot for the exam, it slightly bothers me to have it graded automatically by an AI, particularly if I don’t know the exact criteria used for the grading. They do provide AI feedback, which is nice to figure out what you need to work on, but because it is AI-generated, I am skeptical on its accuracy and truthfulness. Overall, I felt like I put in a lot of effort in the exam only for an automatic grader to spit out a result.</p><p>The other slight complaint I have with the marking scheme is that even if you are able to obtain a flag, only 20% of the grade is given. The other 80% comes from the technique and methodology you used to obtain the flag. I thought this was imbalanced and could be potentially problematic. For instance, if you happen to find the flag really fast because you chose the right technique, you might be still given a low score because you didn’t try other techniques.</p><p>All in all, the trade-off with fast grading is that they need to use automated systems.</p><p>The TryHackMe AI1 certification would be right for you if you want to demonstrate your knowledge and skills in AI Security, and have 2 days to spare to do the exam. Personally, I enjoyed the AI Security learning path more and would recommend you take it even if you decide not to do the certification exam.</p><p>This is just my opinion and experience from when I took the certification. If you had the same experience or thought something different, I would love to hear it in the comments!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e90e23bb9fed" width="1" height="1" alt=""><hr><p><a href="https://meetcyber.net/an-honest-review-of-the-tryhackme-ai1-certification-e90e23bb9fed">An Honest Review of the TryHackMe AI1 Certification</a> was originally published in <a href="https://meetcyber.net/">MeetCyber</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[5 AI Security Certifications Available in 2026]]></title>
            <link>https://pub.towardsai.net/5-ai-security-certifications-available-in-2026-ec6f89a8d9a1?source=rss-5fcb37186b3a------2</link>
            <guid isPermaLink="false">https://medium.com/p/ec6f89a8d9a1</guid>
            <category><![CDATA[ai-security]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[information-security]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Mike Chung]]></dc:creator>
            <pubDate>Sun, 20 Sep 2026 22:01:01 GMT</pubDate>
            <atom:updated>2026-09-20T22:01:01.735Z</atom:updated>
            <content:encoded><![CDATA[<h4>Interested in a career in AI Security? Here are 5 certifications to break into the field.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*H4N1fcgTGPIGvXbVM7AKmA.jpeg" /><figcaption>Photo by Dan Nelson on Unsplash</figcaption></figure><p>As AI is becoming more integrated in our workflows, it’s only a matter of time before AI Security also becomes mainstream. With the recent incident where <a href="https://openai.com/index/hugging-face-incident-and-the-road-ahead/">a rogue agent from Open AI hacked into Hugging Face’s systems</a>, the importance of AI security has only become more and more apparent.</p><p>To get a head start and position yourself for future AI Security roles, here are 5 certifications that you can consider.</p><h3>TryHackMe AI1</h3><p>Link: <a href="https://tryhackme.com/certification/ai-security">https://tryhackme.com/certification/ai-security?vccr=1</a></p><p>TryHackMe is a cybersecurity learning platform that focuses on hands-on exercises and labs. The AI1 certification follows the AI Security learning path in TryHackMe, which consists of 5 modules:</p><ul><li>AI Fundamentals</li><li>Secure AI Systems</li><li>Prompt Security</li><li>AI Supply Chain Security</li><li>Data Poisoning</li></ul><p>The exam is 48 hours long and you will be expected to go through 13 hands-on scenarios involving both attacking and defending AI systems during this time. This is different from traditional certification exams that are usually just multiple-choice questions and answers. TryHackMe has LLMs and AI agents hosted on their platform that you can interact with to get hands-on experience with techniques like prompt injection. The key advantages of the AI1 is that it follows industry standards and mirrors real-world situations. Learning through TryHackMe is also convenient since it is self-paced and flexible so you can learn anywhere, anytime.</p><p>Price-wise, it is $399 to take the AI1 certification exam, and the price includes one free re-take and 3 months of the Premium subscription for TryHackMe.</p><h3>Hack The Box COAE</h3><p>Link: <a href="https://academy.hackthebox.com/preview/certifications/htb-certified-offensive-ai-expert">https://academy.hackthebox.com/preview/certifications/htb-certified-offensive-ai-expert</a></p><p>Hack The Box (HTB) has a hands-on certification called the Certified Offensive AI Expert (COAE). Similar to TryHackMe, HTB is a platform for learning hacking and cybersecurity. To be eligible to take the exam for COAE, the AI Red Teamer job-role path in HTB must first be completed. The curriculum is comprised of 12 modules covering various topics such as adversarial ML, AI data attacks, LLM prompt injection and jailbreaking, evasion techniques and more. The exam itself is a 7-day endeavor where you conduct an AI red team assessment on a simulated enterprise environment hosted on HTB’s infrastructure, culminating in a final report that is submitted for evaluation.</p><p>Since you need to complete the AI Red Teamer job-role path first, you will need to purchase a subscription. The most cost-effective way to get the COAE certification is to get an annual Silver subscription for $490 which includes an exam voucher that is valid for two tries.</p><h3>CompTIA SecAI+</h3><p>Link: <a href="https://www.comptia.org/en/certifications/secai/">https://www.comptia.org/en/certifications/secai/</a></p><p>CompTIA is one of the big players in cybersecurity certifications and recently expanded their offerings to include the SecAI+ certifation. The purpose of the SecAI+ is for practitioners to demonstrate the ability to secure, govern, and integrate AI into cybersecurity operations. According to their exam objective summary, the exam consists of basic AI concepts related to cybersecurity (17%), securing AI systems (40%), AI-assisted security (24%), and AI governance, risk, and compliance (19%). The exam is 60 minutes long and has a maximum of 60 multiple-choice or performance-based questions. The pros of the certification is that is offered by a well-established organization respected by the industry with structured objectives and an organized rubric. To study for the exam, CompTIA has a CertMaster platform that teaches you in digestible videos, readings, and quizzes.</p><p>The current cost for a voucher to take the SecAI+ exam is $298. If you want to bundle it with CertMaster and an exam re-take, it will set you back $628.</p><h3>EC-Council COASP</h3><p>Link: <a href="https://www.eccouncil.org/ai-courses/certified-offensive-ai-security-professional-coasp-north-america/">https://www.eccouncil.org/ai-courses/certified-offensive-ai-security-professional-coasp-north-america/</a></p><p>EC-Council is another big player in the cybersecurity certification space and offers a Certified Offensive AI Security Professional (COASP) certification. This certification is composed of 10 modules:</p><ul><li>Offensive AI and AI System Hacking Methodology</li><li>AI Reconnaissance and Attack Surface Mapping</li><li>AI Vulnerability Scanning and Fuzzing</li><li>Prompt Injection and LLM Application Attacks</li><li>Adversarial Machine Learning and Model Privacy Attacks</li><li>Data and Training Pipeline Attacks</li><li>Agentic AI and Model-to-Model Attacks</li><li>AI Infrastructure and Supply Chain Attacks</li><li>AI Security Testing, Evaluation, and Hardening</li><li>AI Incident Response and Forensics</li></ul><p>The COASP exam is a 6-hour, live-proctored exam with 70 questions total, 65 multiple-choice questions and 5 practical challenges.</p><p>The standalone exam voucher for COASP is $650. For bundles that include the exam voucher with training options, you are looking at $1,699 to $3,499 depending on whether you opt for the on-demand, live online, or in-person course.</p><h3>GIAC Certifications</h3><p>Link: <a href="https://www.giac.org/focus-areas/artificial-intelligence">https://www.giac.org/focus-areas/artificial-intelligence</a></p><p>GIAC develops and administers professional information security certifications, and is the credential and certification affiliate of SANS Institute. It offers the widest range of specialized AI Security certifications, including:</p><ul><li>GIAC AI Penetration Tester (GAIPT)</li><li>GIAC AI Platform Security (GAIPS)</li><li>GIAC AI Security Automation Engineer (GASAE)</li><li>GIAC Offensive AI Analyst (GOAA)</li></ul><p>The assessment is delivered through CyberLive, a hands-on format that uses performance-based challenges in realistic lab environments. CyberLive utilizes virtual machines, real security tools, and authentic code to evaluate real-world capability.</p><p>A standalone exam attempt for GIAC costs $999 and the corresponding course from SANS institute can cost $5,000 — $10,000.</p><h3>Honorable Mentions</h3><p>There are two certifications that deserve an honorable mention. First, is the <strong>ISACA Advanced in AI Security Management (AAISM)</strong>. This certification is an AI-centric security management credential designed to manage AI risks and implement policies. The AAISM is more geared towards information security managers and less technical than the 5 certifications mentioned. The second is <strong>Microsoft’s Cloud and AI Security Engineer Associate</strong>. This is a vendor-specific certification targeting those who work with Microsoft’s products such as Azure and Defender. Both these certifications are also valid options if you are in information security management or work mostly in Microsoft’s ecosystem of products.</p><p>As AI technology becomes more mature, AI Security will also advance, and certifications and training will also catch up to the current state of the technology. As of the writing of this article, ISC2 is also in development of a AI Security certification that is scheduled to be released in late 2026.</p><h3>Which certification should you get?</h3><p>It depends. If you are just starting out in AI Security and want to gain hands-on experience, the TryHackMe AI1 and Hack The Box COAE are your best bet. They are both hands-on assessments that cost less than $500. If you prefer a more knowledge-based and traditional testing format to build up your theoretical foundation, go with CompTIA SecAI+.</p><p>If you are already working in the field of cybersecurity and want to expand to AI Security roles, EC-Council COASP and GIAC certification could also be an option. Since they are significantly more expensive, it would be more worthwhile to have your employer support you during the certification process.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ec6f89a8d9a1" width="1" height="1" alt=""><hr><p><a href="https://pub.towardsai.net/5-ai-security-certifications-available-in-2026-ec6f89a8d9a1">5 AI Security Certifications Available in 2026</a> was originally published in <a href="https://pub.towardsai.net/">Towards AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Self-Hosting Airflow at Home: Automating Stock Price Data Collection]]></title>
            <link>https://pub.towardsai.net/self-hosting-airflow-in-a-home-lab-to-get-up-to-date-stock-ticker-price-data-fff4e57a5243?source=rss-5fcb37186b3a------2</link>
            <guid isPermaLink="false">https://medium.com/p/fff4e57a5243</guid>
            <category><![CDATA[data-engineering]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[apache-airflow]]></category>
            <category><![CDATA[airflow]]></category>
            <dc:creator><![CDATA[Mike Chung]]></dc:creator>
            <pubDate>Sat, 13 Jun 2026 23:52:17 GMT</pubDate>
            <atom:updated>2026-06-18T06:02:18.092Z</atom:updated>
            <content:encoded><![CDATA[<h4>I ran Airflow at home to pull live stock data. Here’s how I did it and what I learned.</h4><p>One of the main goals of creating my home lab is to gain a deeper understanding of Machine Learning Operations (MLOps) and how to productionalize AI workflows. Generally speaking, MLOps and productionalization deals with moving AI models from research into a real-life environment with automation and ability to handle errors gracefully.</p><p>In my previous articles, I set up a <a href="https://medium.com/@fsstance/setting-up-postgresql-in-a-proxmox-homelab-40504aac16dc">PostgreSQL server</a> and an <a href="https://medium.com/@fsstance/setting-up-apache-airflow-in-a-proxmox-container-be3939c3fbc2">Airflow server</a>. These serve as the data foundation for how to get datasets that will be used to train AI models. Now we need to start filling out our PostgreSQL databases with data. We can use Airflow to orchestrate data pipelines so that up-to-date data is loaded into our PostgreSQL database. Setting up this data foundation is typically the first step in the machine learning (ML) process after planning, as you need data to train your models.</p><p>A major reason behind my home lab setup is I want to show that you can self-host the whole ML process with a couple of VMs and containers. Since I have been doing a lot of personal investments lately, let’s work with finance data. With finance data, you can analyze trends, correlate prices, and even try forecasting, making it broadly useful across many scenarios.</p><h3>Configuring Airflow</h3><h4>Running you Airflow Server</h4><p>Before we start, here are things I learnt and implemented that will make life easier with Airflow. When I set up the Airflow server previously, I used the following commands:</p><pre>nohup airflow scheduler &gt; scheduler.log 2&gt;&amp;1 &amp;</pre><pre>nohup airflow dag-processor &gt; dag-processor.log 2&gt;&amp;1 &amp;</pre><pre>nohup airflow triggerer &gt; triggerer.log 2&gt;&amp;1 &amp;</pre><pre>nohup airflow api-server --port 8080 &gt; api-server.log 2&gt;&amp;1 &amp;</pre><p>The<strong> nohup</strong> command allows you to run a process continuously even after you log out or close the shell. The issue I ran into was that the Airflow components would crash and I would have to go to the Airflow server and re-run the commands. Another issue was that starting up all the components required 4 commands which could be annoying to run all the time.</p><p>To solve the latter issue, you can write a script to start or restart Airflow.</p><pre>nano airflow_restart.sh</pre><p>In the airflow_restart.sh file, copy the below code and save it.</p><pre>#!/bin/bash<br>pkill -f &quot;airflow&quot; --ignore-ancestors<br>sleep 2<br><br>echo &quot;Starting scheduler...&quot;<br>nohup airflow scheduler &gt; scheduler.log 2&gt;&amp;1 &amp; echo $! | tee scheduler.pid<br>echo &quot;Starting dag-processor...&quot;<br>nohup airflow dag-processor &gt; dag-processor.log 2&gt;&amp;1 &amp; echo $! | tee dag-processor.pid<br>echo &quot;Starting triggerer...&quot;<br>nohup airflow triggerer &gt; triggerer.log 2&gt;&amp;1 &amp; echo $! | tee triggerer.pid<br>echo &quot;Starting api-server...&quot;<br>nohup airflow api-server --port 8080 &gt; api-server.log 2&gt;&amp;1 &amp; echo $! | tee api-server.pid<br><br>echo &quot;Airflow restarted&quot;</pre><p>This script will kill any processes with “airflow” and restart all 4 components. So instead of running all 4 commands, you can now just run the script.</p><p>To go a step further and solve the issue of having to restart the Airflow components every time it crashes, you can set up Airflow as a system daemon process. A daemon is a background process that typically starts when you boot up your system, restarts when there are failures, and is detached from the shell so you can keep it running at all times.</p><p>First, create a file using:</p><pre>nano /etc/systemd/system/airflow-scheduler.service</pre><p>Next, copy the below text, put it in the file, and save it.</p><pre>[Unit]<br>Description=Airflow Scheduler<br>After=network.target postgresql.service<br>Wants=postgresql.service<br><br>[Service]<br>User=&lt;USER&gt;<br>Group=&lt;GROUP&gt;<br>Environment=PATH=&lt;AIRFLOW_PATH&gt;:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin<br>Environment=AIRFLOW_HOME=&lt;AIRFLOW_FOLDER&gt;<br>ExecStart=&lt;AIRFLOW_PATH&gt; scheduler<br>Restart=on-failure<br>RestartSec=5s<br>StandardOutput=journal<br>StandardError=journal<br><br>[Install]<br>WantedBy=multi-user.target</pre><p>Replace &lt;USER&gt;, &lt;GROUP&gt;, &lt;AIRFLOW_PATH&gt;, and &lt;AIRFLOW_FOLDER&gt; depending on how you set up your Airflow server. Repeat this for all the other components of Airflow: dag-processor, triggerer, api-server.</p><p>Run the below to pick up new system daemon processes.</p><pre>sudo systemctl daemon-reload</pre><p>Then, enable and start all the services.</p><pre># Enable services on boot<br>sudo systemctl enable airflow-scheduler<br>sudo systemctl enable airflow-dag-processor<br>sudo systemctl enable airflow-triggerer<br>sudo systemctl enable airflow-api-server<br><br># Start services<br>sudo systemctl start airflow-scheduler<br>sudo systemctl start airflow-dag-processor<br>sudo systemctl start airflow-triggerer<br>sudo systemctl start airflow-api-server</pre><p>Voila! You now have Airflow running on your server in a more robust way. You can use <strong>systemctl status</strong> to check if each component is running and <strong>journalctl</strong> to see logging information.</p><h4>Adding your PostgreSQL Connection</h4><p>Another thing to do is to set up the PostgreSQL connection in your Airflow server so that it can talk to your PostgreSQL database. Airflow has a handy PostgreSQL connection that lets you specify the connection parameters once, and this allows you to re-use the connection anytime you want to connect to PostgreSQL.</p><p>Edit the <strong>airflow.cfg</strong> file.</p><pre>nano airflow.cfg</pre><p>Find where it says <strong>test_connection</strong> and set it equal to <strong>Enabled</strong>. Restart Airflow (either using the script or restart the daemon processes). This allows you to test your connections through the Airflow UI.</p><p>Now, let’s go to the Airflow UI and add the PostgreSQL connection.</p><p>One the left panel, select <strong>Admin</strong> &gt; <strong>Connections</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/789/1*XvedoZGnIV_piVpAhBBAig.png" /></figure><p>On the top right, select <strong>Add Connection</strong>. Under <strong>Connection Type</strong>, select <strong>Postgres</strong>. Note you may need to run <strong>pip install apache-airflow-providers-postgres </strong>in your Airflow server if you do not see <strong>Postgres</strong> as an option. Fill in your PostgreSQL Host (IP address), Login, Password, Port, and Database.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*m5p3jNJgNR3m2HqRbvq4Cw.png" /></figure><p>You should see your new PostgreSQL connection. To test the connection, click on the graph-like icon. After clicking, it should become like a wi-fi icon. There should also be a green pop-up message on the bottom-right indicating your test connection was successful.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JnSmKnAIHlIGwsmnIaYYZw.png" /></figure><h3>Creating the Airflow Pipeline Code</h3><p>Airflow has been set up and configured properly, so now we can get to writing the code for the Airflow pipeline.</p><p>First, a little bit about Airflow. Airflow works through Directed Acyclic Graphs (DAGs), which is just a term they use to describe structured workflows that can contain multiple data processing tasks. Within a DAG, the tasks can be run in specified orders and with dependencies, so you can set up tasks to run when tasks complete or in parallel with other tasks.</p><p>To get finance data, I used <strong>yfinance</strong>, a python package that leverages Yahoo! Finance’s APIs to get market data. We can use this to extract and write stock ticker price data to our PostgreSQL database.</p><p>Here is the final code for our Airflow DAG which I put in a file named <strong>dag.py</strong>.</p><pre>import pandas as pd<br>import yfinance as yf<br>import os<br>import csv<br>from datetime import date, datetime, timedelta<br>from dateutil.relativedelta import relativedelta<br>from airflow.sdk import DAG<br>from airflow.providers.standard.operators.python import PythonOperator<br>from airflow.providers.postgres.hooks.postgres import PostgresHook<br><br><br>def cad_ticker_price_task():<br>    &quot;&quot;&quot;Fetch 5-year historical price data for Canadian equities and load into PostgreSQL.<br><br>    Reads ticker symbols from watchlist/watchlist_cad.csv, appends the &#39;.TO&#39; suffix<br>    for the Toronto Stock Exchange, downloads OHLCV data via yfinance starting 5 years<br>    ago, and writes the result to the finance.watchlist_cad_ticker_price table using<br>    the Airflow connection pg_data_science_conn. Replaces the table on each run.<br>    &quot;&quot;&quot;<br><br>    # Open csv file to get watchlist and read into list<br>    dag_folder = os.path.dirname(os.path.abspath(__file__))<br>    csv_path = os.path.join(dag_folder, &quot;watchlist/watchlist_cad.csv&quot;)<br>    <br>    with open(csv_path, mode=&quot;r&quot;, encoding=&quot;utf-8&quot;) as file:<br>        csv_reader = csv.reader(file)<br>        watchlist_cad = list(csv_reader)[0]<br>        <br>    watchlist_cad = [x.upper() + &quot;.TO&quot; for x in watchlist_cad]<br><br>    # Create pandas dataframe with data starting from 5 years agi<br>    start_dt = date.today() - relativedelta(years=5)<br><br>    finance_df = yf.download(watchlist_cad, start=start_dt)<br>    finance_df = finance_df.stack(level=1)<br>    finance_df = finance_df.reset_index()<br>    finance_df = finance_df.rename(columns={&#39;level_0&#39;: &#39;Date&#39;})<br>    finance_df.index.name = None<br><br>    # Write dataframe to SQL table<br>    db_hook = PostgresHook(postgres_conn_id=&#39;pg_data_science_conn&#39;)<br>    engine = db_hook.get_sqlalchemy_engine()<br>    table_name = &quot;finance.watchlist_cad_ticker_price&quot;<br><br>    finance_df.to_sql(<br>        name=table_name,<br>        con=engine,<br>        if_exists=&#39;replace&#39;,<br>        index=False,<br>        chunksize=1000<br>    )<br>    print(f&quot;{table_name} written to PostgreSQL&quot;)<br><br><br># Default arguments for Airflow DAG<br>default_args = {<br>    &quot;owner&quot;: &quot;mike&quot;,<br>    &quot;depends_on_past&quot;: False,<br>    &quot;start_date&quot;: datetime(2026, 6, 1),<br>    &quot;retries&quot;: 1,<br>    &quot;retry_delay&quot;: timedelta(minutes=5),<br>}<br><br><br># Airflow DAG<br>with DAG(<br>    dag_id=&quot;finance_pipeline_to_postgres&quot;,<br>    default_args=default_args,<br>    schedule=&quot;@daily&quot;,<br>    catchup=False<br>) as dag:<br>    write_to_db = PythonOperator(<br>        task_id=&quot;watchlist_cad_ticker_price_to_postgres&quot;,<br>        python_callable=cad_ticker_price_task<br>    )<br>    write_to_db</pre><p>In simple terms, the code creates a Airflow DAG named <strong>finance_pipeline_to_postgres</strong> that runs daily. The task takes a list of stock ticker symbols from <strong>watchlist_cad.csv</strong> and extracts stock ticker price data starting from 5 years before the current date via <strong>yfinance</strong>. This is loaded into a pandas dataframe where it is also transformed to the correct structure. The code then connects to the PostgreSQL connection we configured earlier and writes it to a table named <strong>finance.watchlist_cad_ticker_price</strong>.</p><h3>Getting the DAG to Run on the Airflow Server</h3><p>After we have created the <strong>dag.py</strong> file, we need to get it to our Airflow server. By default, on the Airflow server, there is a dedicated <strong>dags</strong> folder where the server looks for Python files to parse and load DAGs. Since I developed the DAG code on a Windows machine separate from the Airflow server, we need to now move it onto the Airflow server into the <strong>dags</strong> folder.</p><p>One simple way is to use <strong>scp</strong>, a command-line utility to transfer files between local and remote machines. If you decide to go this route, you would simply need to run:</p><pre>scp &lt;LOCAL_DAG_PATH&gt;/dag.py &lt;AIRFLOW_USER&gt;@&lt;AIRFLOW_IP&gt;:&lt;AIRFLOW_FOLDER&gt;/dags</pre><p>Here, &lt;LOCAL_DAG_PATH&gt; is where your <strong>dag.py</strong> file is located, &lt;AIRFLOW_USER&gt; and &lt;AIRFLOW_IP&gt; is the username and IP address of your Airflow server, and &lt;AIRFLOW_FOLDER&gt; is the folder on the Airflow server where your <strong>dags</strong> folder is located. After running the command, you should be prompted for the password for your Airflow user, and you will need to input the correct password to transfer the file.</p><p>The problem with the <strong>scp</strong> approach is that you need to run the command every time you make edits to the <strong>dag.py</strong> file. To better mirror production settings, I decided to put the code in a GitHub repository. So, now the development work becomes:</p><ol><li>On my Windows machine, I develop the code for the DAG and push the changes onto the remote GitHub repository.</li><li>On my Airflow server, I then clone the GitHub repository into the <strong>dags</strong> folder. Since Airflow looks iteratively through the folder for Python files, it can find the <strong>dags.py</strong> file and load it.</li><li>Whenever I want to develop the code further, I can make the changes, commit it on my Windows machine, and push it to GitHub. On my Airflow server, I simply pull the new changes.</li></ol><p>From a production point-of-view, this streamlines the process as I can put many files in the GitHub repo in my Windows machine and then pull all the new files and changes into my Airflow server. It also offers version control so I can rollback changes in case things fail.</p><p>To set this up on the Airflow server, we run the following commands.</p><pre># Install Git<br>sudo apt install git<br><br># Clone the remote GitHub repo<br>git clone https://github.com/miketcchung/homelab-data-engineering.git<br><br># Pull changes as needed<br>git pull</pre><p>One thing to note is that on the Airflow server, you will need to have all the required packages already installed to run the <strong>dag.py</strong> code. To fix this, on my Windows machine I wrote a <strong>requirements.txt</strong> file with all the required packages using:</p><pre>pip freeze &gt; requirements.txt</pre><p>Then, on the Airflow server, I pulled the <strong>requirements.txt</strong> file and updated the Python environment with:</p><pre>pip install -r requirements.txt</pre><p>Now that we have our DAG code on our Airflow server and required packages are installed, we just need to make sure that the server reads the DAG correctly.</p><p>Run the following. It should output “No Data Found”, meaning there are no input errors.</p><pre>airflow dags list-import-errors</pre><p>Below are additional commands you can run to ensure the Airflow server is reading the DAG properly.</p><ul><li><strong>airflow dags list</strong> -&gt; shows list of DAGs for the Airflow server, check that your dag_ID is there</li><li><strong>airflow dags unpause dag_ID </strong>-&gt; unpause your DAG if it shows up as paused</li><li><strong>airflow dags trigger dag_ID</strong> -&gt; manually trigger your DAG from the command line</li><li><strong>airflow dags reserialize</strong> -&gt; makes Airflow re-parse all files in the <strong>dags</strong> folder</li></ul><p>In the Airflow UI, you should also be able to see the DAG that you created. On the left panel, select <strong>Dags</strong>. Filter for <strong>Active</strong> DAGs as there are a lot of example DAGs that comes with Airflow.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6N0qdoOcxtSY57Xii18jPg.png" /></figure><p>Verify that it is successful. If it failed, there are logs you can refer to to fix your DAG.</p><p>You now have a working Airflow DAG writing stock ticker price data to your PostgreSQL database.</p><h3>Further Improvements for Airflow in Production</h3><p>You can find the code I used to set up the Airflow DAG and write the stock ticker price dataset to PostgreSQL in this <a href="https://github.com/miketcchung/homelab-data-engineering">GitHub repo</a>. In the future, I plan to put more data engineering code and workflows into this repo, so stay updated for more.</p><p>This is one of my beginning steps toward building production data engineering workflows, and there are additional considerations I came across when developing the code and setting up Airflow.</p><p>One pain point when developing the DAG code is that testing is decoupled from the development process. To test the DAG code, I had to commit changes to GitHub first, and then pull the changes on the Airflow server, trigger the DAG, and monitor the logs for errors. Ideally, there should be a way to test the DAG code on my Windows machine before committing changes to GitHub.</p><p>Another pain point I know will become a larger issue later on is how to manage the Python environments, especially as I scale up Airflow to run multiple DAGs. With the current setup, any time I add a new DAG or edit a DAG to use new packages, I need to install the package on the Airflow server’s Python environment. Python packages have different dependencies, so using the same Python environment for all the DAGs will likely cause conflicts as I scale up. The obvious fix to this is to use Docker operators, which is supported in Airflow. With Docker, we can spin up a container with its own environment for each DAG or task.</p><p>Lastly, it would be nice to have alerting in case DAGs fail, as models depend on the data and would be affected as well. Airflow has built-in alerting for email or Slack, and is something that I will be further developing as the home lab is built out.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fff4e57a5243" width="1" height="1" alt=""><hr><p><a href="https://pub.towardsai.net/self-hosting-airflow-in-a-home-lab-to-get-up-to-date-stock-ticker-price-data-fff4e57a5243">Self-Hosting Airflow at Home: Automating Stock Price Data Collection</a> was originally published in <a href="https://pub.towardsai.net/">Towards AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[OverTheWire Natas 11 -> 12 Walkthrough]]></title>
            <link>https://miketcchung.medium.com/overthewire-natas-11-12-walkthrough-b5f9d47618ec?source=rss-5fcb37186b3a------2</link>
            <guid isPermaLink="false">https://medium.com/p/b5f9d47618ec</guid>
            <category><![CDATA[ctf]]></category>
            <category><![CDATA[web-security]]></category>
            <category><![CDATA[overthewire]]></category>
            <category><![CDATA[natas]]></category>
            <dc:creator><![CDATA[Mike Chung]]></dc:creator>
            <pubDate>Sun, 19 Apr 2026 23:25:31 GMT</pubDate>
            <atom:updated>2026-04-19T23:25:31.930Z</atom:updated>
            <content:encoded><![CDATA[<p>This was one of the Natas levels that I was stuck on for quite some time. Here’s a clear and understandable explanation to help you get through it.</p><h3>Understanding the Problem</h3><p>We are faced with a webpage where you can input and set the background color based on hexadecimal colors.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VRrN7LZ8BSndLU-UYvbgJQ.png" /></figure><p>We are also provided with a hint that “Cookies are protected with XOR encryption”. In Google Chrome, we can find the value of the cookies by right-clicking and selecting the <strong>Inspect</strong> option. A panel should open on the right hand side with the HTML code. At the top, select the <strong>Application</strong> tab. On the left part of the panel, expand <strong>Cookies</strong> and select the only cookie in the choices underneath. You should see a cookie named <strong>data</strong> and its corresponding value. Click on the <strong>data</strong> cookie and look to the bottom. Check the box that says <strong>Show URL-decoded</strong> and you will get the value of the cookie at the bottom.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/738/1*hHWe3m5r66npYNuWchTUeQ.png" /></figure><p>Playing around with the webpage, you’ll notice that if you change the background color input, the value of the cookie also changes, although there are similarities for cookies with similar inputs. With the default input of <strong>#ffffff</strong>, the corresponding cookie is:</p><pre>HmYkBwozJw4WNyAAFyB1VUcqOE1JZjUIBis7ABdmbU1GIjEJAyIxTRg=</pre><p>There is also a link to the source code. Looking at the source code, we see PHP code with three functions: <strong>xor_encrypt</strong>,<strong> loadData</strong>, and <strong>saveData</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/862/1*mWK0B0g3NFf2MTEVJRSdSA.png" /></figure><p>Exploring <strong>xor_encrypt</strong>, it takes an input (<strong>$in</strong>) and a key (<strong>$key</strong>) and XORs them together to produce an output (<strong>$outText</strong>). However, the key is unknown. The main takeaway here is that XOR encryption is easy to break. Even if we don’t know the key, because of the properties of XOR, if we know the input and output and XOR them together, we can get the key.</p><p>Next, <strong>loadData</strong> takes an input called <strong>$def</strong>, sets <strong>$mydata</strong> equal to it, and goes through a bunch of conditional <strong>if </strong>logic. Reading through the logic, the function first decodes and decrypts the cookie and stores it in <strong>$tempdata</strong>. If <strong>$tempdata</strong> is an array, has keys called <strong>showpassword</strong> and <strong>bgcolor</strong>, and <strong>bgcolor</strong> is in the right format, then it sets <strong>$mydata</strong> to the values in <strong>$tempdata</strong>.</p><p>The function <strong>saveData</strong> sets the cookie by first encoding the input (<strong>$d</strong>) as a json string, then using the <strong>xor_encrypt</strong> function, and finally encoding to a base64 value.</p><p>Now, looking through the code that is run, it boils down to this:</p><pre>$defaultdata = array( &quot;showpassword&quot;=&gt;&quot;no&quot;, &quot;bgcolor&quot;=&gt;&quot;#ffffff&quot;);<br><br>$data = loadData($defaultdata);<br><br>if(array_key_exists(&quot;bgcolor&quot;,$_REQUEST)) {<br>    if (preg_match(&#39;/^#(?:[a-f\d]{6})$/i&#39;, $_REQUEST[&#39;bgcolor&#39;])) {<br>        $data[&#39;bgcolor&#39;] = $_REQUEST[&#39;bgcolor&#39;];<br>    }<br>}<br><br>saveData($data);</pre><p>There is a <strong>$defaultdata</strong> variable that has <strong>showpassword</strong> set to “no” and <strong>bgcolor</strong> set to “#ffffff”. Using <strong>loadData</strong> with <strong>$defaultdata</strong> as the argument, a <strong>$data</strong> variable is created. If a properly formatted cookie exists, then <strong>$data</strong> would have its values set to the values in the cookie. If a properly formatted cookies doesn’t exist, <strong>$data</strong> would just be <strong>$defaultdata</strong>.</p><p>The <strong>if</strong> blocks checks if <strong>bgcolor</strong> in <strong>$_REQUEST</strong> exists and is in the right format, and if it is, assigns its <strong>bgcolor</strong> in <strong>$data</strong>. <strong>$_REQUEST</strong> must be what is inputted by the user in the webpage.</p><p>Putting this together, when the webpage is first opened, <strong>$defaultdata</strong> has a bgcolor value of “#ffffff” corresponding to white and <strong>showpassword</strong> equal to “no”. The value isn’t changed by <strong>loadData</strong> because there is no cookie yet, so <strong>$data</strong> would equal <strong>$defaultdata</strong>. The <strong>saveData</strong> function encrypts <strong>$data</strong> as a json string, then uses <strong>xor_encrypt</strong> on it, encodes it to base64, and sets it as the cookie.</p><p>Finally, there is a snippet of code that shows how to get the password for natas12. We need to change <strong>showpassword</strong> in <strong>$data</strong> to “yes”.</p><pre>if($data[&quot;showpassword&quot;] == &quot;yes&quot;) {<br>    print &quot;The password for natas12 is &lt;censored&gt;&lt;br&gt;&quot;;<br>}</pre><h3>Solution</h3><p>Now that we understand the code, we can think through how we can get the password to natas12. The user input doesn’t affect <strong>showpassword</strong>, so we can’t use it to manipulate the PHP variables. Instead, we’ll need to manipulate the cookie to change <strong>$data</strong>. However, the cookie is encrypted with the <strong>xor_encrypt </strong>function which we don’t know the key for, so we’ll need to find the key first.</p><p>We know that for the input</p><pre>$defaultdata = array(&quot;showpassword&quot;=&gt;&quot;no&quot;, &quot;bgcolor&quot;=&gt;&quot;#ffffff&quot;)</pre><p>the corresponding cookie is</p><pre>HmYkBwozJw4WNyAAFyB1VUcqOE1JZjUIBis7ABdmbU1GIjEJAyIxTRg=</pre><p>From our analysis of the code, we also know that the cookie was encoded to base64 after XOR encryption and <strong>$defaultdata</strong> has been encoded to json before XOR encryption. So we will need to encode $defaultdata to json and perform the opposite on the cookie — decode it from base64. Then we can XOR the values to find the key.</p><p>We can write a simple function to do just that. You actually don’t have to use the default input of <strong>#ffffff</strong>, you can choose any input as long as the corresponding cookies is also used. If you don’t have a PHP editor or compiler on your computer like me, you can use <a href="https://onecompiler.com/php">OneCompiler</a>.</p><pre>$defaultdata = array(&quot;showpassword&quot;=&gt;&quot;no&quot;, &quot;bgcolor&quot;=&gt;&quot;#ffffff&quot;);<br>$cookie = &quot;HmYkBwozJw4WNyAAFyB1VUcqOE1JZjUIBis7ABdmbU1GIjEJAyIxTRg=&quot;;<br> <br>function xor_find_key($input, $output) {<br>    $processed_output = base64_decode($output);<br>    $processed_input = json_encode($input);<br>    <br>    $calculated_key = $processed_input ^ $processed_output;<br>     <br>    return $calculated_key;<br>}<br> <br>$key = xor_find_key($defaultdata, $cookie);<br>echo $key;</pre><p>From this function, we find that the key is</p><pre>eDWoeDWoeDWoeDWoeDWoeDWoeDWoeDWoeDWoeDWoe</pre><p>So the key is “edWo” repeated multiple times. If we look back at the <strong>xor_encrypt</strong> function, the loop actually works so that the key is repeated if the input length is longer than the key. So we can simplify the key to “edWo”.</p><p>Now that we know the key, we can creating a variable with <strong>showpassword</strong> equal to “yes” and put it through the <strong>xor_encrypt</strong> function with the key to generate a cookie that will show us the natas12 password.</p><pre>$newdata = array(&quot;showpassword&quot;=&gt;&quot;yes&quot;, &quot;bgcolor&quot;=&gt;&quot;#ffffff&quot;);<br> <br>function xor_encrypt($in) {<br>   $key = &quot;eDWo&quot;;<br>   $text = $in;<br>   $outText = &#39;&#39;;<br><br>   // Iterate through each character<br>   for($i=0;$i&lt;strlen($text);$i++) {<br>     $outText .= $text[$i] ^ $key[$i % strlen($key)];<br>   }<br><br>   return $outText;<br>}<br>  <br>$newcookie = base64_encode(xor_encrypt(json_encode($newdata)));<br>echo $newcookie;</pre><p>This gives us the new cookie we’ll need to show the password.</p><pre>HmYkBwozJw4WNyAAFyB1VUc9MhxHaHUNAic4Awo2dVVHZzEJAyIxCUc5</pre><p>The final part is going into Google Chrome, changing the cookie value, and refreshing the webpage. Using the same <strong>Application </strong>tab, right-click on the cookie value and select <strong>Edit “Value”</strong>. Paste in your new cookie value and press <strong>Enter</strong>. Then refresh the webpage. The natas12 password will now be displayed on the page.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_r9W3m5OR3XRbP_ZPJ9Z2Q.png" /></figure><p>Congratulations! The password for the next level is <strong>yZdkjAYZRd3R7tq7T5kXMjMJlOIkzDeB</strong>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b5f9d47618ec" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Find Tutors for Language Learning]]></title>
            <link>https://miketcchung.medium.com/how-to-find-tutors-for-language-learning-6cb5056f1606?source=rss-5fcb37186b3a------2</link>
            <guid isPermaLink="false">https://medium.com/p/6cb5056f1606</guid>
            <category><![CDATA[tutoring]]></category>
            <category><![CDATA[online-tutoring]]></category>
            <category><![CDATA[tutoring-services]]></category>
            <category><![CDATA[langauge]]></category>
            <category><![CDATA[language-learning]]></category>
            <dc:creator><![CDATA[Mike Chung]]></dc:creator>
            <pubDate>Wed, 15 Apr 2026 21:58:41 GMT</pubDate>
            <atom:updated>2026-04-15T21:58:41.034Z</atom:updated>
            <content:encoded><![CDATA[<p>A difficult part of learning a new language is finding people to speak the language with and learn from. Especially if you live in a country where the language is not regularly spoken, it can be challenging to find opportunities to practice the language. Luckily, we live in a world connected by the internet, so you can find people right from the convenience of your computer.</p><h3>Tutoring</h3><p>At the beginning of your language learning journey, it’s helpful to find a tutor. Native and experienced speakers can walk you through all the little nuances of the language you want to learn. With how interconnected the world is now, you can easily find a native speaker tutor from another country at affordable prices. Below are a few of the platforms that connect you with tutors.</p><h4>italki</h4><p>Link: <a href="https://www.italki.com/en">https://www.italki.com/en</a></p><p>Price: $25–60 per hour, trial for $10 per 30 minutes</p><p>Through italki, you can connect with tutors around the world. After selecting the language you want to learn, you can select from a list of tutors. Each tutor has a short 1–2 minute video introducing themself and a description of themselves including their background, teaching style, availability and pricing. You can also see their rating, how many lessons they’ve completed, and whether they are a Community Tutor (more focused on casual conversations) or a Professional Teacher (generally certified/licensed and more academic-focused).</p><p>You can also book a trial lesson with the tutor you want at a very reasonable price. I booked a 30 minute lesson with a Japanese Community Tutor for about $10, which is a steal. My experience with the trial lessons were positive, with the tutors being kind, friendly and helpful. They even sent me a list of words and phrases after our lesson. If you take lesson packages with a tutor, they’ll create a lesson plan with you according to your goals. Prices usually range between $25-$60 per hour depending on the popularity and expertise of your tutor.</p><h4>AmazingTalker</h4><p>Link: <a href="https://ca.amazingtalker.com/">https://ca.amazingtalker.com/</a></p><p>Price: $20–45 per 50 minutes, trial as low as $2 per 25 minutes.</p><p>Similar to italki, although the prices are even more affordable. Their interface is more traditional with a search bar followed by filters for country, price, availabity and etc. Looking through the French tutors, you can get a trial lesson for less than $5 and a full 50 minute lesson for around than $20. You can also see each tutor’s rating, lessons completed, background and introduction video. There’s also an option for Instant Tutoring, which means you can potentially start the lesson 30 minutes after booking!</p><p>Popular languages on the platform include Spanish, Korean, Japanese, French, and Chinese. There is also a function on the website where they can match you with a tutor based on a set of questions you answer. The set of questions includes the student’s age, learning goal, time availability, and budget.</p><h4>Preply</h4><p>Link: <a href="https://preply.com/">https://preply.com/</a></p><p>Price: $25–70 per hour</p><p>Preply is another online platform for finding an online tutor. In this platform, you first answer a set of questions such as your learning goal, your current level in that language, what country you want your tutor to be from and what languages you want them to speak. It then matches you with a list of tutors based on your answers. Preply would be best-suited for finding a personalized tutor, especially if you have certain criteria for the tutor that you want.</p><p>As with the other platforms, tutors have a short intro video and written introduction. You can also see detailed reviews on the tutor’s reassurance, clarity, progress, and preparation. Tutors who are certified also get a “Professional” tag and those who are highly reviewed get a “Super Tutor” tag. Prices are slightly higher than other platforms, although you get free trial lessons upon registration.</p><h4>Superprof</h4><p>Link: <a href="https://www.superprof.ca/">https://www.superprof.ca/</a></p><p>Price: $30–60 per hour</p><p>For those who prefer in-person lessons, you can try Superprof. By inputting the language you want to learn and your location, you can find tutors nearby to help on your language learning journey. Depending on the tutor, classes can be both online or face-to-face. If you choose the face-to-face option, you can discuss with your tutor whether you would like to have the lesson at the tutor’s house or if you would like the tutor to visit you or meet at a public place. For exceptional tutors, they are given the “Super Prof” or “Ambasssador” badge. You can view information on the tutor’s page, including a bio, lesson format, rates, and reviews and recommendations from other students.</p><p>Generally, tutors offer the first 30-minute lesson free. Some tutors may charge additional fees for travel over a certain distance.</p><p>Happy learning!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6cb5056f1606" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[HomeLab Series #3. Setting up Apache Airflow in a Proxmox Container]]></title>
            <link>https://miketcchung.medium.com/setting-up-apache-airflow-in-a-proxmox-container-be3939c3fbc2?source=rss-5fcb37186b3a------2</link>
            <guid isPermaLink="false">https://medium.com/p/be3939c3fbc2</guid>
            <category><![CDATA[data]]></category>
            <category><![CDATA[data-engineering]]></category>
            <category><![CDATA[homelab]]></category>
            <category><![CDATA[proxmox]]></category>
            <category><![CDATA[airflow]]></category>
            <dc:creator><![CDATA[Mike Chung]]></dc:creator>
            <pubDate>Wed, 25 Feb 2026 07:30:56 GMT</pubDate>
            <atom:updated>2026-03-06T17:33:24.227Z</atom:updated>
            <content:encoded><![CDATA[<p>Previously, we set up and configured <a href="https://medium.com/@fsstance/setting-up-postgresql-in-a-proxmox-homelab-40504aac16dc">PostgreSQL on an Ubuntu Server in our Proxmox HomeLab</a>. Let’s continue building upon the data infrastructure in our HomeLab.</p><p>The tool that we want to implement next is <strong>Apache Airflow</strong>, a platform for data engineering. Data engineering deals with data pipelines so you can schedule and automatically update the tables and databases that you have on your system. In the real world, data is not stationary and you always want to do your analyses and build your models with the newest data available. Moreover, if you have raw data that needs to be cleaned and processed, you can set up data pipelines to do this regularly and get the dataset that you desire.</p><p>Let’s learn how to set up Airflow in our HomeLab. Before going through this tutorial, you need to already have a Proxmox Virtual Environment set up for hosting your virtual machines and containers. Also, PostgreSQL is recommended, although Airflow does allow an option that uses SQLite if you do not have PostgreSQL available.</p><p>We will be setting up Airflow on a Linux Container (LXC). We chose a LXC over a Virtual Machine (VM) because they are more lightweight and portable. They share the host machine’s operating system, so you don’t have to install one like you have to do in a VM. This also makes LXCs less resource-intensive and faster to start up than a VM. This makes it ideal for Airflow as we want something isolated that is easy to scale up if needed.</p><h3>Linux Container Creation</h3><p>First, you will need to download a template. In one of your Proxmox storages, click “CT Templates” and then “Templates”. You will be given a list of container templates that you can download onto your storage.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/899/1*qYsnIgT2lfsAnqVb0quCmQ.png" /></figure><p>Select <strong>ubuntu-24.04-standard </strong>and click “Download”. Give it a few minutes to download and it will eventually say “TASK OK” in the output, which means that it has finished downloading.</p><p>Now, let’s create our LXC. On the top right in Proxmox, click “Create CT”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/721/1*-1CxYmYiJMeIx4e3e_841w.png" /></figure><p>Fill out the tabs as follows:</p><ul><li>In “General”, set the “Hostname”. I set mine as “airflow-ubuntu-lxc”. Also, set a secure password in “Password”.</li><li>In “Template”, navigate to the storage you downloaded the template from and select the ubuntu-24.04-standard template that you downloaded.</li><li>Under “Disks”, set the disk size to 50 GB.</li><li>For “CPU”, allocate 2 cores.</li><li>For “Memory”, allocate 4 GB of “Memory” and leave “Swap” at 512 MB.</li><li>In “Network”, select “DHCP” for both IPv4 and IPv6.</li><li>In “DNS”, leave it as the default configurations.</li><li>Click “Confirm” to create the LXC.</li></ul><p>You should see “TASK OK” in the output confirming that the Ubuntu LXC was successfully created.</p><h3>Airflow Installation</h3><p>Now, start up your newly created airflow-ubuntu-lxc and open the console. You will first be prompted to login. The default username is “root” and the password is the one you set when creating the Ubuntu LXC.</p><p>First, run the command below to ensure all your packages are up-to-date.</p><pre>sudo apt update &amp;&amp; sudo apt upgrade</pre><p>*Note: It is recommended that you take a snapshot of your LXC before installation in case anything goes wrong. That way, if something breaks, you won’t have to delete and create the VM again.</p><p>We will be installing Airflow through Python, so let’s first install Python and its environment and package manager.</p><pre>sudo apt-get install python3.12 python3-pip python3-venv</pre><p>To confirm that Python is installed correctly, run the command below.</p><pre>python3 --version</pre><p>It should show that you have Python 3.12 installed.</p><p>Next, we’ll set the environment variable that will define where the Airflow configuration files will be stored.</p><p>Run the following to edit the file that stores your environment variables.</p><pre>sudo nano /etc/environment</pre><p>Add in the following line to the bottom of the file. Feel free to change the location if you want to store the files somewhere else.</p><pre>AIRFLOW_HOME=&quot;~/airflow&quot;</pre><p>Now, logout and login back into the root account for the changes to take place. You can also run <strong>echo </strong>on your environment variable to confirm that it is set correctly.</p><p>We need to first make a Python virtual environment before installing Airflow. The virtual environment helps to isolate the packages that are installed from other projects and prevent conflict from dependencies.</p><pre>python3 -m venv airflow-env</pre><p>Then, activate your virtual environment.</p><pre>source airflow-env/bin/activate</pre><p>Now, we install Airflow using <strong>pip</strong>. We need to make sure that we install it with constraints, as installing without constraints may cause the installation to throw an error.</p><pre>pip3 install &quot;apache-airflow==3.1.7&quot; --constraint &quot;https://raw.githubusercontent.com/apache/airflow/constraints-3.1.7/constraints-3.12.txt&quot;</pre><p>Voila! We now have Airflow installed on our Ubuntu LXC.</p><h3>Optional: Setting up a PostgreSQL database and connecting it to Airflow</h3><p>This section is optional, as Airflow will use SQLite as the database if you do not have PostgreSQL available. SQLite stores your database locally and can thus be more limited in its capabilities. You can create the SQLite database by running:</p><pre>airflow db init</pre><p>If you want a more production-like platform, it is recommended to use PostgreSQL or another client-server database system. If you want to set up PostgreSQL on Proxmox, you can also refer to <a href="https://fsstance.medium.com/setting-up-postgresql-in-a-proxmox-homelab-40504aac16dc">my previous article</a>.</p><p>Open up a terminal in your PostgreSQL server. Login to your postgres account or any other superuser.</p><pre>su postgres</pre><p>Now, run <strong>psql</strong> to enter into the interface for PostgreSQL. Use the below SQL commands to create a user and database. Feel free to change the username, password, and database name according to your liking.</p><pre>CREATE DATABASE airflow_db;<br>CREATE USER airflow_user WITH PASSWORD &#39;airflow_pass&#39;;<br><br>GRANT ALL PRIVILEGES ON DATABASE airflow_db TO airflow_user;<br>GRANT ALL ON SCHEMA public TO airflow_user;</pre><p>Go back to your Airflow Ubuntu LXC terminal. We will need to edit the <strong>airflow.cfg</strong> file to point toward our PostgreSQL database.</p><pre>sudo nano airflow/airflow.cfg</pre><p>Scroll down to where it mentions <strong>sql_alchemy_conn</strong>. It should currently equal to <strong>sqlite:////root/airflow/airflow.db</strong>. Change this to be equal to</p><pre>postgresql+psycopg2://airflow_user:airflow_pass@&lt;POSTGRES_IP:POSTGRES_PORT&gt;/airflow_db</pre><p>where <strong>&lt;POSTGRES_IP:POSTGRES_PORT&gt;</strong> is the IP address and port of the PostgreSQL server.</p><p>We’ll also need to install a few libraries that are needed for the connection from Airflow to PostgreSQL. First, run the following:</p><pre>sudo apt install libpq-dev python3-dev</pre><p>Next, install <strong>psycopg2</strong> and <strong>asyncpg</strong>.</p><pre>pip3 install psycopg2 asyncpg</pre><p>To confirm successful installation of the libraries, use the below command.</p><pre>airflow config get-value database sql_alchemy_conn</pre><p>At the bottom of the output, you should see the value you set in <strong>sql_alchemy_conn </strong>in the <strong>airflow.cfg</strong> file.</p><p>To test the database connection we can run the below.</p><pre>airflow db check</pre><p>The output should say the connection is successful.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/789/1*_VdTfP7yiFoRvNCpsnz93w.png" /></figure><h3>Creating an Airflow user</h3><p>Before we start using Airflow, we need to create a Airflow user that will allow us to login and interact with the web interface. The version that we installed, Airflow 3.X, by default uses <strong>SimpleAuthManager</strong>, an authentication manager used for development and testing. We’ll need to change this to another authentication manager that allows us to create users.</p><p>First, run the below command to install the new authentication manager.</p><pre>pip3 install apache-airflow-providers-fab</pre><p>Now, we need to set the new authentication manager in our <strong>airflow.cfg</strong> file. Run</p><pre>sudo nano airflow/airflow.cfg</pre><p>and find the line where it mentions <strong>auth_manager</strong>. Change it to be equal to:</p><pre>airflow.providers.fab.auth_manager.fab_auth_manager.FabAuthManager</pre><p>After this is done, you should be able to create users in Airflow. Run the command below to create a user with the parameters of your choosing.</p><pre>airflow users create \<br>  --username admin<br>  --password adminpassword<br>  --firstname Tom<br>  --lastname Hardy<br>  --role Admin<br>  --email tom.hardy@gmail.com</pre><p>At the end of the output you should see <strong>User “admin” created with role “Admin”</strong>.</p><h3>Running Airflow and accessing the Web Interface</h3><p>There are two ways to run Airflow. You can use</p><pre>airflow standalone</pre><p>command to intialize the database, create a user, and start all components. A more controlled way is to run each component separately using <strong>nohup</strong>.</p><pre>nohup airflow scheduler &gt; scheduler.log 2&gt;&amp;1 &amp;</pre><pre>nohup airflow dag-processor &gt; dag-processor.log 2&gt;&amp;1 &amp;</pre><pre>nohup airflow triggerer &gt; triggerer.log 2&gt;&amp;1 &amp;</pre><pre>nohup airflow api-server --port 8080 &gt; api-server.log 2&gt;&amp;1 &amp;</pre><p>Amazing! You now have Airflow running in your HomeLab.</p><p>Now, you can go into a machine on your network and access the web interface. Open up a web browser and navigate to <strong>http://&lt;AIRFLOW_IP&gt;:8080</strong> where <strong>&lt;AIRFLOW_IP&gt;</strong> is the IP address of your Ubuntu LXC. You can run <strong>ip addr</strong> on the terminal of your LXC if you don’t know the IP address. You should see the Airflow login page once you navigate to this address.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yu68bfVPLoghcDWGCiiBgw.png" /></figure><p>Login with the credentials that you set when you created the Airflow user. If you followed my instructions, the username and password should be “admin” and “adminpassword”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9Di3E32YO5urUECnGo2gWQ.png" /></figure><p>If everything is successful, you should see the welcome page for Airflow. On the welcome page, you should see checkmarks for <strong>MetaDatabase</strong>, <strong>Scheduler</strong>, <strong>Triggerer</strong>, and <strong>Dag Processor </strong>indicating they are up and running. If you don’t see these checkmarks, go back to your LXC terminal and re-run the command for the component.</p><h4>References</h4><ul><li><a href="https://airflow.apache.org/docs/">Documentation</a></li><li><a href="https://airflow.apache.org/docs/apache-airflow/stable/installation/installing-from-pypi.html">Installation from PyPI</a></li><li><a href="https://airflow.apache.org/docs/apache-airflow/stable/installation/setting-up-the-database.html">Setting up the database</a></li><li><a href="https://airflow.apache.org/docs/apache-airflow/stable/howto/set-up-database.html">Set up a Database Backend</a></li><li><a href="https://pypi.org/project/apache-airflow/">apache-airflow</a></li><li><a href="https://pypi.org/project/psycopg2/">psycopg2</a></li><li><a href="https://stackoverflow.com/questions/35104097/how-to-install-psycopg2-with-pg-config-error">How to install psycopg2 with pg_config error?</a></li><li><a href="https://github.com/apache/airflow/issues/51304">Airflow v3 user creation command fails · Issue #51304 · apache/airflow</a></li><li><a href="https://stackoverflow.com/questions/79596338/airflow-users-create-command-not-working-with-3-0-version-failing">airflow users create command not working with 3.0 version-failing</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=be3939c3fbc2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Supercharge your Language Learning with AI and Technology]]></title>
            <link>https://miketcchung.medium.com/supercharge-your-language-learning-with-ai-and-technology-0cf456638514?source=rss-5fcb37186b3a------2</link>
            <guid isPermaLink="false">https://medium.com/p/0cf456638514</guid>
            <category><![CDATA[language]]></category>
            <category><![CDATA[language-translation]]></category>
            <category><![CDATA[language-learning]]></category>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[transcription]]></category>
            <dc:creator><![CDATA[Mike Chung]]></dc:creator>
            <pubDate>Thu, 08 Jan 2026 04:58:03 GMT</pubDate>
            <atom:updated>2026-01-08T04:58:03.025Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L727ustUl5I7wd_VW9JUBA.jpeg" /><figcaption>Photo by Ling App on Unsplash.</figcaption></figure><p>Work smarter not harder is the new mantra of the technological age.</p><p>Throughout my language learning process, I realized that <strong>the key to learning any language effectively is consistency</strong>. According to psychology research, it takes 5 to 7 repetitions for information to move from short-term memory to long-term memory. In practice, these repetitions need to be spaced apart in time so in can be best incorporated into your memory. The problem with this is that, although we know consistency is key, we struggle with motivation and finding the time to learn the new language long-term.</p><p>I was the type that tried to separate a few hours a week to study a new language. I quickly ran into major obstacles. Try reading a textbook for a hour and you’ll slowly find that your eyes start to glaze over. There are certain studious types who may find this approach doable, but I’m guessing for the vast majority of the population, we get bored pretty quick. Not de-emphasizing the importance of having a structured and systematic education — I definitely recommend you spend time reviewing study material if it’s in your capacity. But in the end, <strong>you need to find something that you enjoy and works for you</strong>.</p><p>The other blocker is that <strong>a few hours a week simply won’t cut it</strong> if you’re seeking to become fluent. Ideally, you need to review everyday or the next time you come back to learn, you’ll be reviewing what you forgot. This eventually becomes a cycle where you’re learning simply to maintain your language ability, not for improvement.</p><p>Eventually, I found out that <strong>to be consistent, you need to immerse yourself in the language</strong>. And not only through textbooks and study materials, but through everyday things. If you’re in a country where the main language is the one that you are studying, then it’s easy to be immersed. You have many opportunities in your everyday functions to practice and be exposed to the new language, whether it’s by listening to people talking, trying to read signs and advertisements, or talking with friends and acquaintances. But what if you’re trying to learn a foreign language from another country? Luckily, we live in an interconnected world where you can access other countries’ TV, news, podcasts, books and other media from the Internet. Depending on your interest, you can find something you like that’s in the language you would like to learn. Although it’s hard to read a textbook for an hour, we can comfortably watch a three hour movie or a multi-series TV show. The more you enjoy doing something, the more likely you are to continue. So, the ultimate takeaway is to <strong>find activities you enjoy doing in the language you want to learn and immerse yourself in it to achieve long-term consistency</strong>.</p><p>Technological advances have made language learning content more readily available. There are websites where you can post questions and answers in language learning communities, connect to native speaker tutors, and make friends across the world. AI tools can now generate images and articles, convert speech to text, and provide real-time translation. With these advances, it’s becoming effortless to immerse yourself in another language and achieve the fluency that you desire.</p><p>As an avid language learner (currently learning Japanese, Mandarin Chinese, and Taiwanese), I want to share a comprehensive list of the websites and tools that I’ve tried and use on a regular basis. Also, I want to share how I use these websites and tools in the hope that what worked for me, might also work for one of you all.</p><p>Without further ado, let’s get into it.</p><h3>Transcription/Translation</h3><p>The first range of tools pertain to transcribing audio and/or translating text to a target language.</p><h4>Language Reactor (Google Chrome Extension)</h4><p>Link: <a href="https://chromewebstore.google.com/detail/language-reactor/hoombieeljmmljlkjmnheibnpciblicm?hl=en">https://chromewebstore.google.com/detail/language-reactor/hoombieeljmmljlkjmnheibnpciblicm?hl=en</a></p><p>Who doesn’t love to watch TV? I’m sure we all can spend many hours binge-watching a series. Language Reactor is a free browser extension that translates subtitles from one language to another on websites like Netflix and YouTube.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*px-264iGi3wj1IIFBcQbeA.png" /><figcaption>Image from Godzilla Minus One, directed by Takashi Yamazaki, 2023, via Netflix.</figcaption></figure><p>This handy tool shows two language subtitles on top of each other so you can watch a movie in the language you want to learn. The tool also allows you to highlight the subtitle words while you’re watching and see the translations. Highlighting the text also automatically pauses the video so you can study the translated words. Clicking on the word opens up an in-depth translation with many examples of how the word is used. If you have a Pro subscription, you can also add words to a vocabulary list that you can export for studying later on.</p><p>Since Netflix shows usually have multiple languages available, the movies and series that you watch will usually have human translations available. There is also an option to use machine translations if human translations are not available. Usually, when I’m starting to learn a new language, I watch a movie in the new language with subtitles in both the new language and English. Once I’ve obtained a certain level of competency, I start to challenge myself and turn off the English subtitles. At this point, I can recognize most of the words in the new language, but if I come across a word I don’t know, I can use the highlight function to get the translation quickly and continue watching the show.</p><p>Happy watching!</p><h4>Google Translate (Google Chrome Extension)</h4><p>Link: <a href="https://chromewebstore.google.com/detail/google-translate/aapbdbdomjkkjkaonfhkkikfgjllcleb?hl=en">https://chromewebstore.google.com/detail/google-translate/aapbdbdomjkkjkaonfhkkikfgjllcleb?hl=en</a></p><p>Not only is Google Translate a website, it’s also an extension that you can download and install onto your browser. The extension allows you to translate words, phrases, and even large bodies of text by simply highlighting what you want to translate.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_nkhQ8FCB4ONSU075zMp0g.png" /><figcaption>Screenshot of <a href="https://tw.news.yahoo.com/%E4%B8%8D%E8%AB%8B%E6%B1%82%E5%86%8D%E5%AF%A9-%E8%94%A1%E6%AD%A3%E5%85%83%E6%9B%9D%E5%85%A5%E7%9B%A3%E7%8D%84%E5%89%8D%E5%BF%83%E9%A1%98-090516295.html">news article from Yahoo News Taiwan</a>.</figcaption></figure><p>I typically use this tool when reading news or articles. Starting off with easier articles, you can easily progress and build up your vocabulary. Soon enough, you’ll find yourself reading longer and more difficult articles without the help of translation! If you’re not into news, you can also read blogs, books, forums, or whatever else you want.</p><h4>Audio Transcribers</h4><p>TurboScribe: <a href="https://turboscribe.ai/">https://turboscribe.ai/</a></p><p>Evernote AI Transcribe: <a href="https://evernote.com/ai-transcribe">https://evernote.com/ai-transcribe</a></p><p>Another idea that can boost your language learning is to find a podcast in the language that you would like to learn. Download the podcast as an audio file. Then, use an AI audio transcriber to transcribe the audio into text. Pop the text transcription into Google Translate to get an English translation. Listen to the podcast while following along with the transcription and English translation.</p><p>I found this especially helpful when trying to learn the language’s casual and conversational nuances. Textbooks and study materials generally teach the language from an academic and formal setting which sometimes does not translate to real-world communication. For languages like Japanese where conversations are drastically different than what’s seen in the textbook, this has been a game-changer.</p><p>One thing to note is that AI audio transcribers can sometimes be incorrect. Particularly in conversational podcasts where it can be messy with people talking over each other, mumbling, muttering, or speaking in accents. Translating the transcriptions can also generate some errors as well, since not everything can be directly translated from one language to another. That being said, AI has come a long way and has become more efficient in providing error-free transcriptions and translations.</p><p>To download podcasts as audio files, I had to use a bit of computer programming knowledge to create a script that downloaded the file from <a href="https://analytics.podtrac.com/">Podtrac</a>. You can also try websites like <a href="https://podcasttomp3.com/">PodcastToMp3</a> or <a href="https://beamly.com/tools/apple-podcasts-downloader/">beamly</a> if you don’t want to mess with programming.</p><p>For those looking for free AI audio transcribers, you can do a trial with <a href="https://turboscribe.ai/">TurboScribe</a>. There are limits to the free version in that only 30 minutes of the audio file can be transcribed. However, it generates a decent transcription with punctuation included. <a href="https://evernote.com/ai-transcribe">Evernote</a> also has a AI transcribe tool, but it doesn’t have punctuation, so it’s more difficult to follow without the periods and commas.</p><p>If you’re great with writing programs and coding, you can also use <a href="https://openai.com/index/whisper/">Whisper</a> by OpenAI, the same company that developed ChatGPT. In about half a day, I was able to write a quick script that took an audio file and generated a transcript in a text file. However, it did not include punctuation, so in the end, I ended up just using TurboScribe.</p><h3>Flashcards</h3><p>Once you’ve committed a bunch of vocabulary to memory, you’ll find that you need to continuously review previous words while adding new words to memory. My favorite way to do this is using flashcards.</p><h4>Brainscape</h4><p>Link: <a href="https://www.brainscape.com/landing">https://www.brainscape.com/landing</a></p><p>Brainscape is the tool that I found most simple to create flashcards and practice with them. First of all, you need a list of vocabulary words and their definitions/translations. I made a list by compiling words that I didn’t know from news, TV, movies, podcasts, books and other sources, and putting them into an Excel file. Then, I made a class on Brainscape and imported the list to create decks of 30 vocabulary words each.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bc6hIgDoHqNMKOdbURgw7A.png" /><figcaption>Screenshot of Brainscape’s interface for flashcards.</figcaption></figure><p>Voila! Now you can study flashcards anytime. You can also download the app on your phone so you can study on the go, such as when you’re taking public transit or waiting in line.</p><p>A key advantage of using Brainscape is that they use cognitive learning principles such as spaced repetition so the vocabulary can be efficiently incorporated and retained in memory. After each flashcard, you can rate on a scale of 1 to 5 how well you know the vocabulary and it will cleverly determine the perfect interval on when to show the flashcard again. It’s recommended to study these flashcards in frequent, bite-sized sessions, so they can be retained in your memory. Since it’s available on most devices, it’s easy to find time to practice and learn.</p><p>Making and practicing with flashcards that you created on Brainscape is free. However, there is also a marketplace where creators make flashcards and put them in classes. You can usually access the first few flashcard decks of a class by these creators for free, but if you want access to their full class, you would need to either contact them or get a Pro subscription for $7.99 per month.</p><p>If you want to try it out and are learning Mandarin Chinese, here’s a link to a class that I’ve created for Traditional Chinese: <a href="https://www.brainscape.com/packs/traditional-chinese-vocabulary-with-zhuyin-bopomofo-23479181">https://www.brainscape.com/packs/traditional-chinese-vocabulary-with-zhuyin-bopomofo-23479181</a>.</p><h4>Anki</h4><p>Link: <a href="https://apps.ankiweb.net/">https://apps.ankiweb.net/</a></p><p>Anki is another flashcard app that is similar to Brainscape. It also has smart features that let you rate how well you know words and helps you focus on more challenging words. The main difference is that Anki predominantly functions through a program that you download onto your computer. It does also have a web version, but you would need to download the program first.</p><p>Anki does have a mobile app as well, although it costs $24.99 on the Apple Store. One reason you might want to use Anki is that it offers more features for customization. It includes media support so you can add audio, images and video to your flashcards. It also allows you to change the layout and timing of the flashcards. There’s also add-ons you can install to extend the functionality.</p><p>Overall, if you’re into customized flashcards with rich functionality, go with Anki. If you prefer something more simple and out-of-the-box, go with Brainscape.</p><h3>Chatbots</h3><p>The last area I want to bring up are AI chatbots. Since the release of ChatGPT in 2022, generative AI has been advancing at a rapid pace. Currently, ChatGPT supports a vast array of languages, although it’s performance in less common languages may be questionable. You can talk to ChatGPT in the language you want to learn to quickly get relevant content or language learning help. I don’t recommend you rely on this as your main source of language learning though, as chatbots can get things wrong and sound unnatural. In the end, it’s best to find a real-life language learning partner or community rather than talk to a robot.</p><p>If you’re looking for something more advanced, you can look into chatbots or generative AI systems built specifically for language learning. These systems are trained on language content so, theoretically, they should be less error-prone and have additional functionalities. Below are options with links that you can check out.</p><h4>Copycat Cafe</h4><p>Link: <a href="https://copycatcafe.com/">https://copycatcafe.com/</a></p><p>Copycat Cafe is an app that assists in French language learning. They have an AI assistant named Copy that provides conversation practice and also scores your pronunciation. If you’re a shy person and are embarrassed to make mistakes in front of other people, the AI assistant could be suitable for you. However, the app is quite pricey, coming in at $40.20 per month or $241.23 per year.</p><h4>Langua Talk</h4><p>Link: <a href="https://languatalk.com/">https://languatalk.com/</a></p><p>Langua has a human-like AI language coach that helps you practice your language. The main selling point is their AI sounds natural, authentic, and human-like because it is trained on speech from native speakers. Langua also provides personalized feedback, such as written corrections with explanations, verbal coaching, post-chat reports and audio summaries based on your conversations with their AI language coach. Depending on if you want a monthly or annual subscription, you can get a 5 day or 7 day free trial. If you want to get the subscription, you have two choices: the Standard for $16.67 per month or Unlimited for $25 per month. Currently, there are 20+ languages that are supported including English, French, and Spanish.</p><h4>MemBot</h4><p>Link: <a href="https://www.memrise.com/blog/introducing-membot">https://www.memrise.com/</a></p><p>MemBot is offered by Memrise. Using GPT-3, the same technology ChatGPT is built upon, MemBot was created as an AI language partner for people to practice speaking a foreign language before entering the wild. Ideally, MemBot should be used as a stepping stone, as practicing with people is still the number one way for language proficiency. MemBot currently supports over 30 languages including Chinese and Japanese. The subscription costs $19.99 monthly, $59.43 annually, or $144.99 for your lifetime.</p><p>AI and technology has made language learning more accessible, more convenient and more efficient. Try out one of these tools and feel free to connect to let me know how it goes!</p><p><strong>Disclaimer: </strong>I am an Authorized Creator on Brainscape, but I am not sponsored by any of the companies mentioned in this article. Opinions reflect mine only.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0cf456638514" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[HomeLab Series #2. Setting up PostgreSQL in Proxmox]]></title>
            <link>https://miketcchung.medium.com/setting-up-postgresql-in-a-proxmox-homelab-40504aac16dc?source=rss-5fcb37186b3a------2</link>
            <guid isPermaLink="false">https://medium.com/p/40504aac16dc</guid>
            <category><![CDATA[homelab]]></category>
            <category><![CDATA[rdbms]]></category>
            <category><![CDATA[postgresql]]></category>
            <category><![CDATA[proxmox]]></category>
            <category><![CDATA[database]]></category>
            <dc:creator><![CDATA[Mike Chung]]></dc:creator>
            <pubDate>Sun, 14 Dec 2025 20:25:57 GMT</pubDate>
            <atom:updated>2026-03-06T17:39:24.685Z</atom:updated>
            <content:encoded><![CDATA[<p>Before building that amazing AI project, you need data. Although you can work with data in .csv files, you’ll soon find that there’s limitations in the size and types of data you can access.</p><p>From an organization’s standpoint, as the amount of your data scales up, you also want to have a solution that can handle multiple datasets, address relationships between the data, and store backups in case things go awry.</p><p>Enter <strong>PostgreSQL</strong>, an open-source relational database management system (RDBMS). The “relational” means that the data is organized into structured tables with rows and columns, and the tables can be related to each other through “keys” or similar columns between datasets. With PostgreSQL, you can create, read, update, and delete rows, columns, and even entire tables. To read the data, you can use SQL to query and combine data into the dataset that you want.</p><p>Let’s see how we can set up these PostgreSQL on our HomeLab. This tool will serve as the foundation for our data and future AI projects.</p><h3><strong>Virtual Machine (VM) Setup</strong></h3><p>In my current HomeLab setup, I have Proxmox installed on a Mini PC server. This is hooked up via Ethernet to my internet modem and router so it’s accessible on my local network. I do most of my work on a Acer Nitro 5 laptop with Windows 11 Home as the Operating System (OS).</p><p>To create the VM, go into Proxmox and click “Create VM” on the top right.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/717/1*lpoXl7Q0uG19DsyH2Vd41Q.png" /></figure><p>In the resulting menu, the tabs are filled out as follows.</p><ul><li>In the “General” tab, set the “Name”. I set mine as “postgresql-ubuntu”.</li><li>Under “OS”, I navigated to an ISO for Ubuntu Server 24.04 that I put on Proxmox. The ISO can be downloaded from this link: <a href="https://ubuntu.com/download/server">https://ubuntu.com/download/server</a>.</li><li>In “System”, leave the default configurations as is.</li><li>In “Disks”, set the disk size to 500 GB.</li><li>For “CPU”, allocate 4 cores.</li><li>For “Memory”, allocate 4 GB of RAM.</li><li>In the “Network” tab, leave the default configurations as is.</li></ul><p>Depending on your HomeLab setup, you may want to tweak the configurations, such as adding more or less storage by changing the disk size. You may also want to adjust the CPU and Memory if you expect to run your compute through PostgreSQL. This would make run queries faster and allow you to cache more data into memory.</p><p>PostgreSQL runs on most major operating systems including well-known versions of Linux. In this tutorial, we use Ubuntu Server because it is open-source, stable, and well-supported. You can also use Debian, Red Hat, or SUSE if that’s what you’re more familiar with.</p><h3>Ubuntu Server Installation</h3><p>Now, let’s fire up our newly created Ubuntu Server VM and open up the console. The first time you open the VM, it will go through the setup to install the OS. Follow the below steps:</p><ul><li>Select “English” as your language.</li><li>If it says “Installer update available”, select “Continue without updating”.</li><li>Choose “Ubuntu Server” as the base for installation.</li><li>Leave “Network configuration” as default and “Proxy configuration” as blank.</li><li>Leave everything else as the default configuration and click “Done”.</li></ul><p>You will also be asked to configure your profile. This is to identify the system and create initial login credentials. I set it as:</p><ul><li>Name: mike</li><li>Servers Name: postgresql-ubuntu</li><li>username: postgres-user</li><li>password: postgres-pass</li></ul><p>If you’re asked to upgrade to Ubuntu Pro, you can click “Skip”.</p><p>After completing the setup steps, the OS will begin installation. It may take a while so feel free to take a break and come back to it. Once it’s complete, you’ll get a message saying “Installation complete!”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Svt7JWmZp0veHFtbci22Rw.png" /></figure><p>Select “Reboot Now”. After your system starts up, you’ll be asked to login. Use the username and password that you created during the setup. Congraulations! You’ve successfully created a Ubuntu Server VM.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/677/1*KCpVwyjV9yjXt9e7-pXjqA.png" /></figure><h3>PostgreSQL Installation</h3><p>Now, let’s set up PostgreSQL on the Ubuntu Server VM so we can start working with databases.</p><p>First, let’s run this command to get the latest packages and install it onto our VM.</p><pre>sudo apt update &amp;&amp; sudo apt upgrade</pre><p>Then, we can install PostgreSQL as a package with the following command.</p><pre>sudo apt install postgresql</pre><p>At the time this article was written, PostgreSQL 16 was installed, which we can confirm with the below command.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/455/1*6ZK-Fub9OnVmjKnYflv53Q.png" /></figure><p>Next, we use <strong>systemctl</strong> to start PostgreSQL and confirm that it’s running.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/772/1*iALH9yk_36t9JQuVtwzRzA.png" /></figure><p>When you install PostgreSQL, a user called <strong>postgres</strong> is created by default in your Linux system and in PostgreSQL. To start using PostgreSQL, you need to have a user in Linux and in PostgreSQL with the same username. Now, let’s logon to our <strong>postgres</strong> user.</p><pre>su postgres</pre><p>We can then access PostgreSQL using <strong>psql</strong>. The command line should change to the following.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/493/1*v_RiIkXXT6XjJHDBYHUy9w.png" /></figure><p>The <strong>postgres</strong> user doesn’t have a password by default so let’s create a password for it with the command below (replace &lt;password&gt; with your password).</p><pre>ALTER USER postgres PASSWORD &#39;&lt;password&gt;&#39;;</pre><p>It’s not good security practice to use the default user, so let’s create another account called <strong>user1</strong>.</p><pre>CREATE ROLE user1 WITH LOGIN PASSWORD &#39;&lt;password&gt;&#39;;</pre><p>Let’s also create a database for our new user and assign it as the owner.</p><pre>CREATE DATABASE user1_db OWNER user1;</pre><p>We can confirm the user is created with <strong>\du</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/597/1*NEMnkFxt00NtxwSTank0hA.png" /></figure><p>We can also confirm the database is created and assigned to <strong>user1 </strong>with <strong>\l</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LUXptF19EYruc13a8ImY8A.png" /></figure><p>To login to our new user, first exit psql with <strong>\q </strong>and then run:</p><pre>psql -h localhost -U user1 -d user1_db</pre><h3>Configuring PostgreSQL for Remote Connection</h3><p>The next thing we want to do is configure PostgreSQL so we can access it on other machines in our network. In our HomeLab, we want to do most of our projects and work on our host Windows laptop and just use PostgreSQL for storing and managing datasets.</p><p>To allow connections from our network, we need to first edit the <strong>postgresql.conf</strong> file. This can be found at <strong>/etc/postgresql/16/main/postgresql.conf </strong>(you may need to change the 16 if you have a different version number). By default, when PostgreSQL is installed, it only allows connection from the local machine. To change this, we un-comment the <strong>listen_addresses</strong> line in <strong>postgresql.conf </strong>and change it to:</p><pre>listen_addresses = &#39;*&#39;</pre><p>This enables systems from any IP address to connect to your server. You can also limit it to just your network if you don’t want to expose the server to everyone.</p><p>Next, we need to edit the <strong>pg_hba.conf</strong> file found at <strong>/etc/postgresql/16/main/pg_hba.conf</strong>. This file controls client authentication. Scroll down to where it says</p><pre>#TYPE  DATABASE  USER  ADDRESS METHOD</pre><p>and add this below.</p><pre>host  all  all  all  scram-sha-256</pre><p>This allows all IP addresses to connect through any user to any database in your PostgreSQL server. The scram-sha-256 refers to the authentication method, and is generally secure for most purposes.</p><p>Finally, let’s restart PostgreSQL so the changes can take effect.</p><pre>sudo systemctl restart postgresql</pre><p>We can test the remote connection from our Windows laptop. You can do this in two ways — through <strong>psql</strong> or <strong>pgAdmin</strong>. To use psql in Windows, you need to download the Windows version of PostgreSQL (<a href="https://www.postgresql.org/download/windows/">https://www.postgresql.org/download/windows/</a>). A simpler way to interact with PostgreSQL if you are not familiar with the command line is pgadmin (<a href="https://www.pgadmin.org/download/">https://www.pgadmin.org/download/</a>), a free interface for managing PostgreSQL.</p><p>I’ll show you how to test the connection with pgadmin. Download and install the latest version of pgAdmin (pgAdmin 4 in my case). During installation, you can just use the default options.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BmiDpv0eN3gMPUJEUD9Enw.png" /></figure><p>Follow the below steps:</p><ul><li>Select “Add New Server”.</li><li>In the “General” tab, type in the name. I put “postgresql-ubuntu”.</li><li>In “Connection”, put in the IP address of your PostgreSQL server. This can be found by running <strong>ip addr</strong> on your Ubuntu Server VM. For “Username”, you can use <strong>postgres</strong> or <strong>user1</strong>. Put the password of the account in “Password”. Also, toggle on “Save password?”.</li><li>Leave everything else as default.</li></ul><p>Click “Save”. If your connect is successful, you should be able to see your PostgreSQL server under “Servers” in the left panel. Congratulation! You now have PostgreSQL set up in your HomeLab.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yyAc-__jmgU-dqNeiPNfPw.png" /></figure><p>The next step is to migrate your data into PostgreSQL. The great thing about PostgreSQL and relational databases is that there are many ways to read and write data. You can do so directly through pgAdmin 4, through SQL, or even other programming languages such as Python.</p><p>Hope this article helps you and see you all next time!</p><h3>References</h3><ol><li><a href="https://documentation.ubuntu.com/server/how-to/databases/install-postgresql/">https://documentation.ubuntu.com/server/how-to/databases/install-postgresql/</a></li><li><a href="https://www.digitalocean.com/community/tutorials/how-to-install-postgresql-on-ubuntu-20-04-quickstart">https://www.digitalocean.com/community/tutorials/how-to-install-postgresql-on-ubuntu-20-04-quickstart</a></li><li><a href="https://neon.com/postgresql/postgresql-getting-started/install-postgresql-linux">https://neon.com/postgresql/postgresql-getting-started/install-postgresql-linux</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=40504aac16dc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[HomeLab Series #1. Creating a HomeLab for Cybersecurity and AI]]></title>
            <link>https://miketcchung.medium.com/planning-a-homelab-for-cybersecurity-and-ai-667728ed1e73?source=rss-5fcb37186b3a------2</link>
            <guid isPermaLink="false">https://medium.com/p/667728ed1e73</guid>
            <category><![CDATA[proxmox]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[homelab]]></category>
            <category><![CDATA[information-technology]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Mike Chung]]></dc:creator>
            <pubDate>Fri, 22 Aug 2025 00:29:43 GMT</pubDate>
            <atom:updated>2026-03-06T17:34:56.812Z</atom:updated>
            <content:encoded><![CDATA[<p>For those who don’t know, a HomeLab is a setup that you make at home for learning and experimenting with technology. Although companies have mostly switched to the Cloud in recent years, it’s a fun and worthwhile experience to start from first principles and understand how to self-host various kinds of servers for different business needs.</p><p>My background is in Data Science and Cybersecurity, so my HomeLab will also naturally tend toward these areas. However, a lot of the initial setup requires working with hardware and networks, so I will go over a few of these concepts as well. Let’s get started!</p><p><strong>Hardware</strong></p><p>The first thing I did was to purchase a Mini PC to use as a server. I opted for a MINISFORUM NAB9 Mini PC purchased from Amazon for around $750 CAD. These are the specs:</p><ul><li>Intel Core i9–12900HK processor with 20 CPUs</li><li>Intel Iris Xe graphics card</li><li>32 GB RAM</li><li>2 x 1 TB SSD hard drive (one of them dug up from an old laptop)</li></ul><p>The Mini PC was hooked up to the internet router through an ethernet connection. And that’s all the hardware you need! Another alternative is to purchase an old server that businesses often sell once they upgrade their architecture, but according to forums and discussions with my IT friends, it’s much more efficiency energy-wise to just use a Mini PC. Not to mention that servers to make a lot of noise so you would have to put it somewhere where it wouldn’t disrupt your everyday life.</p><p>Besides the Mini PC, there’s a few things you can also purchase to make your HomeLab a bit nicer and your life a little easier. I dug up an old mouse and keyboard that I used for initially setting up the Mini PC and also bought a portable monitor on Amazon for $100 CAD.</p><p><strong>Setting up Proxmox</strong></p><p><a href="https://www.proxmox.com/en/">Proxmox</a> is what we call a hypervisor, a platform for virtualization where we can create many handy things like virtual networks, virtual machines, containers and etc.</p><p>To set up Proxmox, I essentially followed the instructions in this <a href="https://www.youtube.com/watch?v=kqZNFD0JNBc&amp;t=64s&amp;ab_channel=DigitalMirror">YouTube video by Digital Mirror</a>. First, I created a bootable USB drive on my laptop using <a href="https://rufus.ie/en/">Rufus</a>. Then I plugged this USB drive onto the Mini PC before booting, and installed Proxmox according to the video. This is where the portable monitor, mouse and keyboard came in handy, as I’m not sure how you could set it up if you weren’t able to interact with the Mini PC.</p><p>When it was all finished, I was able to navigate to the IP address on my browser and access the home screen for Proxmox.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Zv0OYBiMYIdBinH4jUqbzg.png" /></figure><p>Proxmox is the foundation of your HomeLab because it allows you to create multiple servers on one piece of hardware. There’s a bunch of online resources showing what Proxmox can do and I’ve only scratched the surface in my current HomeLab setup.</p><p><strong>Creating a OpenMediaVault NAS</strong></p><p>The first thing I decided to create was a network-attached storage (NAS). I have computers, laptops and other devices connected through my router, so it would be amazing if there weas shared storage that all of these devices could access. I could then use this storage for large folders and files so my laptop’s hard drives can be cleared out for extra space. I could also use the shared storage for backups that take up a lot of space. This is exactly what a NAS solves so let’s get into it!</p><p>There are a few options out there such as TrueNAS and Unraid, but I decided to go with <a href="https://www.openmediavault.org/">OpenMediaVault</a> because of it’s simplicity and out-of-box setup. I also only had two hard drives so I wasn’t planning to implement redundant disks at the moment. In a business setting, having redundant storage is definitely the way to go because you don’t want a crash or accidental deletion to destroy your business.</p><p>For this setup, I relied again on a <a href="https://www.youtube.com/watch?v=ivtKb_FMq50&amp;ab_channel=BarmineTech">YouTube video by Barmine Tech</a>. I’m glad everything can be found on YouTube nowadays. I downloaded the OpenMediaVault ISO file into my Proxmox storage and then created a Virtual Machine (VM) with the ISO. I used two hard drives for the VM: one with 32GB for the operating system and one with 500GB for the actual storage. I also gave it 3GB of RAM and kept the network settings at default.</p><p>Now we boot up the newly created VM and get OpenMediaVault installed on the system.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*etOu1k5H_3AcMb_zaFKvww.png" /></figure><p>There’s quite a few items that you can configure on OpenMediaVault. As a starting point, it’s good practice to:</p><ul><li>Change the default password for administrator</li><li>Create another user for configuration management</li><li>Download and install all updates</li></ul><p>Once all the setup and configuration has been done, you should be able to map the shared folder onto the drives of your devices. I moved all my photos and videos onto the shared folder since they were taking up a lot of space on my hard drives.</p><p>One thing I didn’t do yet is create multiple users and configure the permissions so that some folders are accessible to certain folders while others are not. All the devices on my network currently use the same user. In an industry setting, best practice would be to use the principle of least privilege so users can only access what they need to. I just wanted to set up a NAS easily and quickly for now, but this is something I’ll play around with and explore later on in my HomeLab journey.</p><p><strong>Additional VMs</strong></p><p>I tinkered around with Proxmox more and created more VMs for areas that I’m interested in. I won’t go too much into detail for each VM, but this is something I’ll look into explaining in a later article.</p><ul><li><a href="https://www.kali.org/get-kali/">Kali Desktop</a>: Kali is an open-source Linux OS designed for ethical hacking. It comes pre-installed with tools for network mapping, penetration testing, security testing and etc.</li><li><a href="https://www.parrotsec.org/">Parrot OS Desktop</a>: Parrot OS is another open-source Linux OS that is widely used by information security professionals. While Kali is more focused on offensive security, Parrot OS has an intuitive interface and many privacy-focused tools.</li><li><a href="https://www.zabbix.com/">Zabbix</a> Server: I spun up a VM with Ubuntu Server as the OS and installed Zabbix. Zabbix is a useful software tool to monitor the various devices in your network. By installing a Zabbix agent on each device, the agent will send performance data to the server. On the server, you can customize dashboards according to how you want observe your network and IT architecture. The setup for Zabbix was guided by <a href="https://www.learntohomelab.com/homelabseries/EP0_startguide/">Learn To HomeLab</a>.</li><li><a href="https://www.postgresql.org/">PostgreSQL</a> Server: A large part of AI depends on data, so I set up a PostgreSQL database to store data that I might eventually need for training or fine-tuning AI systems. This would also be for educational purposes on how relational databases work and how they could potentially be exploited.</li><li><a href="https://tailscale.com/">Tailscale</a> Server: Tailscale provides a mesh virtual private network (VPN) so you can connect to your local network as long as your device is connected to the internet. What this means is that I can access all my VMs when I’m on the go, including the shared folders on my NAS.</li></ul><p>Thanks for following my journey as I build a HomeLab for Cybersecurity and AI! I hope this shows that starting a HomeLab does not require a ton of resources and it’s easy to find tutorials, videos, and blog posts on how to configure and troubleshoot the various tools you want to implement. Stay tuned for more articles and content!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=667728ed1e73" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Artificial Intelligence and Machine Learning Applications in Forensics Sciences]]></title>
            <link>https://miketcchung.medium.com/artificial-intelligence-and-machine-learning-applications-in-forensics-sciences-358af4c84f35?source=rss-5fcb37186b3a------2</link>
            <guid isPermaLink="false">https://medium.com/p/358af4c84f35</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[forensics]]></category>
            <dc:creator><![CDATA[Mike Chung]]></dc:creator>
            <pubDate>Fri, 16 Aug 2024 21:21:34 GMT</pubDate>
            <atom:updated>2024-08-16T21:21:34.664Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*7hNcvodwP-H005LT" /><figcaption>Photo by <a href="https://unsplash.com/@tinkerman?utm_source=medium&amp;utm_medium=referral">Immo Wegmann</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The future is moving forward, and Artificial Intelligence (AI) and Machine Learning (ML) is at the forefront. Industries and organizations all around the globe have been beginning to adopt AI and ML, bringing in transformations in workflows and processes.</p><p>But what about applications in forensics sciences? No doubt, there are numerous ethical and regulatory considerations in this field, as there should be as it affects the rest of the life of a person. Nevertheless, there have been projects popping up on innovative ways people have applied AI and ML to discover the truth and apprehend criminals.</p><p>Here are some interesting applications that I found.</p><p><strong>Catching Pedophiles with Image Recognition</strong></p><p>VICE released an <a href="https://www.youtube.com/watch?v=N0vbYBWzNMk">video</a> on using AI to identify people based on vein patterns on the back of their hands. From this, they could catch pedophiles who often record themselves performing heinous acts with their faces covered but hands visible. It was claimed that they could say with 100% certainty that, if the veins on the back of the hands do not match, the person was not the same. As of the making of the video, they have not found any two vein patterns, freckle patterns, or skin crease patterns that have matched. In a possibly landmark case, this evidence was heard in court in a case where a girl alleged her father sexually abused her. This has been used to help police secure over 40 life sentences in the UK.</p><p>This is still actively being researched by Lancaster University in a 5-year project called <a href="https://wp.lancs.ac.uk/h-unique/">H-Unique</a>. Their core data comes from citizens donating photos of their hands and filling out a short questionnaire.</p><p>Although there is not much detail on how they extract the vein patterns and compare images, I guess that they apply the same algorithms as those used in facial recognition. Potential algorithms that could have been used include Eigenface, Principal Component Analysis (PCA), or Convolutional Neural Network (CNN) approaches.</p><p><strong>ML in Digital Forensics</strong></p><p>As our lives become more intertwined with technology, cybersecurity becomes a important factor in ensuring our safety and privacy. Digital forensics serves to analyze the footprints that criminals leave when conducting malicious activity in the cyber world.</p><p>Besides common applications like filtering spam mail and recognizing suspicious activity in surveillance videos, AI and ML can also be for more complex use cases such as network and malware forensics. Some areas in network forensics include labeling malicious traffic, identifying DDoS attacks, and classifying attack behavior. For malware, AI and ML can assist in detecting and responding to malware. Also, with new AI tools at our disposal to generate falsified content such as deepfake images or videos, AI and ML can help detect manipulated photos and videos to prevent the spread of disinformation.</p><p>A <a href="https://www.mdpi.com/2076-3417/13/18/10169#:~:text=Machine%20Learning%20(ML),and%20automate%20DF%20investigative%20procedures.">research paper</a> also highlighted the use of machine learning techniques in digital forensics for smart environments where various devices and appliances are connected to the internet. Since the amount of data collected by smart devices is immense, ML can be used to scan the source code of applications, analyze tracking logs, and detect suspicious files. This was put into practice to build an intelligent intrusion detection system to combat cyber attacks. By automating these processes, we can create a more secure ecosystem of smart devices.</p><p><strong>ML in DNA Profiling</strong></p><p>Now that the processing and analysis of complex, unstructured data is possible, many avenues for forensic DNA analysis are opened.</p><p>In a <a href="https://www.sciencedirect.com/science/article/pii/S1872497323001692">research paper</a>, authors describe in detail how ML can be used to analyze biological data to find DNA and RNA markers. These markers can be utilized to confirm the origin and estimate the age of the sample. Other applications include predicting the ancestry which is more of a forensic intelligence application. ML can also add value to DNA evidence, such as allowing the possibility of estimating the interval post-mortem (after death) or reporting on its source and activity level.</p><p>Finally, with the advent of generative models, research is being undertaken on other use cases such as artificial aging of an individual’s photograph or age prediction from a facial image, which could be theoretically be used as evidence in court. Another exciting application is it may be possible in the future to predict physical traits such as eye color, hair color, skin color and even face morphology from a DAN sample. Taking it one step further, generative models could even simulate some images of potential perpetrators from a DNA sample alone.</p><p>What is in the future of AI and ML in forensics sciences? In my opinion, these are a few keys that lie in wait.</p><p><strong>AI ethics will be critical for bringing projects to the real-world. </strong>It is well-known that models can be biased and cause unfair outcomes. One of the most famous examples is the <a href="https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing">Correctional Offender Management Profiling for Alternative Sanctions (COMPAS)</a>. This model was essentially a tool for corrections to assess the risk of defendants sentenced to a crime. However, the model predictions were eventually found to discriminate against black defendants, wrongly labeling them as high risk at almost twice the rate of white defendants. Another case where AI failed has been in <a href="https://www.technologyreview.com/2020/07/17/1005396/predictive-policing-algorithms-racist-dismantled-machine-learning-bias-criminal-justice/">predictive policing</a>. Since Black people are more likely to be arrested, predictive policing resulted in a disproportionate of police response to Black communities.</p><p><strong>Regulations will need to be in place to effectively implement AI and ML in forensics science. </strong>As a starting point, balanced datasets that are not biased towards any one race should be necessary for any models that are used in forensics science. Model explainability is also crucial since findings will need to be communicated to a judge or jury. Many algorithms are still a black box so significant development will be required so a lay-person will be able to understand how a model makes its predictions.</p><p>Lastly, any decisions made for humans should not be automated or dependant on AI alone. Instead, <strong>AI and ML should be used to assist forensics professionals with their decision-making. </strong>As more cases arise with AI and ML evidence, courts will need to set precedents and make judgments on the validity and acceptance of the evidence. This would lead to certain techniques being admitted while others being throw out.</p><p>I am excited at what the future will bring and will certainly be on the lookout for cases where AI and ML are influencing forensics sciences. Hopefully, these examples have also made you more inspired to tackle projects in this field!</p><p><strong>References</strong></p><p>The Use of Machine Learning in Digital Forensics: Review Paper. <a href="https://www.atlantis-press.com/article/125984186.pdf">https://www.atlantis-press.com/article/125984186.pdf</a></p><p>Machine-Learning Forensics: State of the Art in the Use of Machine-Learning Techniques for Digital Forensics Investigations within Smart Environments. <a href="https://www.mdpi.com/2076-3417/13/18/10169#:~:text=Machine%20Learning%20(ML),and%20automate%20DF%20investigative%20procedures.">https://www.mdpi.com/2076-3417/13/18/10169#:~:text=Machine%20Learning%20(ML),and%20automate%20DF%20investigative%20procedures.</a></p><p>Machine learning applications in forensic DNA profiling: A critical review. <a href="https://www.sciencedirect.com/science/article/pii/S1872497323001692">https://www.sciencedirect.com/science/article/pii/S1872497323001692</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=358af4c84f35" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>