<?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 Lijie Zhou on Medium]]></title>
        <description><![CDATA[Stories by Lijie Zhou on Medium]]></description>
        <link>https://medium.com/@lijie-zhou?source=rss-e44b1724aa04------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*YDqrJCoQ3UBxdVb2VGBvBw.jpeg</url>
            <title>Stories by Lijie Zhou on Medium</title>
            <link>https://medium.com/@lijie-zhou?source=rss-e44b1724aa04------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 05:37:17 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@lijie-zhou/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[What five years of running a small foundation taught me]]></title>
            <link>https://lijie-zhou.medium.com/what-five-years-of-running-a-small-foundation-taught-me-8ff76b8d0d84?source=rss-e44b1724aa04------2</link>
            <guid isPermaLink="false">https://medium.com/p/8ff76b8d0d84</guid>
            <category><![CDATA[education]]></category>
            <category><![CDATA[nonprofit]]></category>
            <dc:creator><![CDATA[Lijie Zhou]]></dc:creator>
            <pubDate>Mon, 08 Jun 2026 22:08:34 GMT</pubDate>
            <atom:updated>2026-06-08T22:30:00.791Z</atom:updated>
            <content:encoded><![CDATA[<p>In January 2021, I was working from home in the Bay Area, one of thousands of engineers on Zoom calls while the world tried to figure out what came next. The company I worked at had just gone through a round of layoffs. My old professor from SFSU emailed asking if we had openings for new grads. His students needed a foothold, and the usual paths had closed. Around the same time, Chuck (my husband and a university professor) was watching the same disparity play out in his own classroom: some students joining lectures from family yachts, others sharing a bedroom with three siblings, everyone technically “in class” but living in completely different worlds. The pandemic didn’t land the same way for everyone, and that gap was impossible to unsee. Chuck and I kept coming back to it, in conversations over dinner, on walks, alongside everything else we were doing. Eventually those conversations became a decision: to start a <a href="https://www.zhoueesleyfoundation.com/">nonprofit</a> and direct our energy toward communities we felt a deep connection to.</p><h3>Why an operating foundation, not a grant-making one</h3><p>The foundation’s first university partnership was with my own alma mater, <a href="https://www.zhoueesleyfoundation.com/programs/sfsu/">San Francisco State University</a>. That wasn’t a coincidence. When I was a student there, I majored in education, not computer science. Breaking into tech meant pushing against a lot of doubt, including my own. It was a professor at SFSU who gave me the first real opening, and I’ve never forgotten what that felt like: one person deciding you were worth a chance.</p><p>Lijie Zhou is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.</p><p>So when Chuck and I started thinking about what the foundation should actually do, the answer came from that experience. We didn’t want to write checks and hope something good happened downstream. We wanted to be in the room, designing the program, training the mentors, talking to the students. That’s what an operating foundation makes possible.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*sFAcLIAhwU5aLSyN.jpeg" /><figcaption>Me with my friends at the SFSU graduation ceremony in 2018.</figcaption></figure><p>Our first program at SFSU was called 0/1. We recruited over a hundred students and matched them with more than ten alumni mentors, focused specifically on interview preparation, the exact moment where a career can open or close. Running that first cohort taught us things we didn’t expect: students needed career guidance and referrals just as much as mock interviews; peer support mattered as much as mentor time, so we created a student ambassador role in the next cohort. And one of the students we worked with got a full-time offer at Google, then came back the following year to mentor the next group. That loop is exactly what we were trying to build.</p><h3>What it’s been like in practice</h3><p>People sometimes ask how I fit this alongside full-time engineering jobs. The honest answer is that we don’t keep it separate; we’ve built it into our life and how we travel. Our first volunteer teaching trip was during our <a href="https://www.zhoueesleyfoundation.com/programs/taiwan/">honeymoon in Taiwan</a>. We called it a “volunteer-moon,” and it turned out to set the tone for everything that followed. Trips became programs. Vacations became residencies. It stopped feeling like a sacrifice and started feeling like just how we spend our time together.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*7FugofBFJa0jVP30.jpeg" /><figcaption>Students we taught from Taiwan wrote up personal notes to thank us.</figcaption></figure><p>What that looks like in practice varies a lot by place. In Molokai (the least developed of Hawaii’s main islands) we’ve taught <a href="https://www.zhoueesleyfoundation.com/programs/molokai/">cybersecurity to high schoolers</a> and co-led AI and entrepreneurship sessions at Kaunakakai Elementary alongside teacher Kawika Gonzales, designing the curriculum around the island’s own community values rather than importing a mainland template. In Penang, we ran <a href="https://www.zhoueesleyfoundation.com/programs/malaysia/">a teacher training cohort </a>with the Penang Science Cluster: roughly 25 teachers who, across their own classrooms, reached about 2,500 students. <a href="https://newsletter.chuckeesley.com/p/why-were-betting-more-on-teachers">The multiplier is the whole point</a>. If you can change what a teacher believes is possible, you reach everyone they ever teach. And in <a href="https://www.zhoueesleyfoundation.com/programs/tanzania/">Tanzania in 2023</a>, we traveled to LOHADA, an orphanage, through a connection James Juma (one of Chuck’s former students) had built over years. We taught alongside Abisola, a Stanford PhD alum who co-taught the curriculum. While we were there, we learned the orphanage had been trying to acquire a tractor. <a href="https://youtu.be/n6KZpIPb2ao?si=5TJa-0dMjJQ5f5_n">We bought a local painting, ran an auction back home, and contributed the proceeds. </a>It raised $1,854. It wasn’t curriculum work, but it was what the people we were there to serve actually needed.</p><h3>What surprised me about the partnership side</h3><p>The foundation also holds a small <a href="https://www.zhoueesleyfoundation.com/impact/portfolio/">portfolio of mission-related investments</a>, what we call MRI. The idea came from a pattern we kept noticing: startups genuinely built around social impact often struggle to raise from traditional VCs, whose incentive structures push toward scale and exit timelines that don’t always fit a mission-driven company. We wanted to be a different kind of capital.</p><p>One of those companies is <a href="https://empohealth.com/">Empo Health</a>, which makes an in-home imaging device to help patients and podiatrists catch diabetic foot ulcers early, before infection sets in, before amputation becomes the conversation. We invested because the mission was personal: we lost a family member to complications from a diabetic foot infection. Empo has faced everything a hard-to-fund company faces, supply chain disruptions during COVID, a lengthy FDA review process, the ordinary brutality of early-stage healthcare. We stayed because we believe in what they’re building, and because we’ve watched them prove, slowly, that they’re heading somewhere real.</p><p>What surprised me most from this work isn’t the companies themselves. It’s the awareness gap. Mission-driven founders don’t usually know that private foundations are even an option. They default to chasing name-brand VCs for validation and network, without realizing that the misalignment between their mission and traditional return expectations is structural, not just a fundraising problem. Part of what the foundation can do is simply show up and tell them there’s another path.</p><p>I’ve also learned things about leadership that my engineering career never taught me. Sitting on a board means evaluating strategy, not just executing it, asking whether the opportunity is real, whether the team can get there, where the leverage actually is. I’m involved actively, not just as a check-writer. That shift in perspective has changed how I think about my own work too.</p><h3>What five years has taught me about my own story</h3><p>Five years in, I’m more convinced than ever that what we set out to do matters, and more aware of how much harder it is than it looks from the outside.</p><p>The world has changed faster than we expected. AI is reshaping what it means to have access to tools, to capital, to education. In some ways that’s the foundation’s thesis accelerating: more people can build things now than could five years ago. But I’ve also become more insistent about a question that gets lost in the excitement: what are we building, and for whom? Efficiency is not the only optimizer. The same technology that opens doors can also narrow them, depending on whose values shape it. Running the foundation has made that concrete for me in a way that working in tech alone never did.</p><p>It’s also taught me that access and agency look different everywhere. In Tanzania, students don’t have computers. We taught basic CS concepts with chalk on a blackboard, and the students were completely engaged, hungry for it. That surprised me, and it recalibrated something. The theory travels. The implementation has to be rebuilt from scratch in every context. You can’t parachute in with a curriculum and assume it fits.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*e64TiFz-RYcokKRd.jpeg" /><figcaption>James Juma taught his Stanford E145 Entrepreneurship to students at LOHADA</figcaption></figure><p>What I understand about my own story now that I didn’t in 2021 is harder to articulate, but I’ll try. I came into tech as an outsider, wrong major, wrong background, too many people who quietly signaled I didn’t belong. The foundation has been, among other things, a way of testing what I actually believe about that experience: whether one person’s intervention can change a trajectory, whether the feeling of belonging can be deliberately created, whether any of this compounds. The students have given me more clarity about my own path than I expected. Watching someone move from doubt to possibility, the way I once did, is the closest thing I’ve found to an answer.</p><p><em>Originally published at </em><a href="https://open.substack.com/pub/zhoulijie/p/how-to-rule-the-world?r=3g0q3&amp;utm_campaign=post-expanded-share&amp;utm_medium=web"><em>https://zhoulijie.substack.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8ff76b8d0d84" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to rule the world?]]></title>
            <link>https://lijie-zhou.medium.com/how-to-rule-the-world-b1b3b0232c91?source=rss-e44b1724aa04------2</link>
            <guid isPermaLink="false">https://medium.com/p/b1b3b0232c91</guid>
            <category><![CDATA[education]]></category>
            <category><![CDATA[philantrophy]]></category>
            <dc:creator><![CDATA[Lijie Zhou]]></dc:creator>
            <pubDate>Sat, 23 May 2026 21:33:09 GMT</pubDate>
            <atom:updated>2026-05-23T21:33:09.384Z</atom:updated>
            <content:encoded><![CDATA[<h4>On elite access, and what it’s actually for.</h4><p>Theo Baker’s <a href="https://theobaker.info/">recent book</a> about his Stanford undergrad years asks a question that people outside Silicon Valley have been wondering about for a long time: is elite education really about learning, or is it about access to networks, to capital, to power? Baker, who broke the story that led to <a href="https://www.npr.org/2023/07/19/1188828810/stanford-university-president-resigns">Stanford president Marc Tessier-Lavigne’s resignation</a> and who was <a href="https://techcrunch.com/2026/05/18/theo-baker-spent-four-years-investigating-stanford-before-he-leaves-heres-what-he-found/">courted with caviar dinners by VCs</a> hoping to tap his Stanford network has a pretty good vantage point to ask.</p><p>Chuck and I have been part of this community for over fifteen years. A lot of what Baker describes is true. <strong>But the more interesting question, to me, isn’t whether elite access is real: it obviously is. It’s what that access is for.</strong></p><h3>Two worlds in the Bay Area</h3><p>When I moved to the Bay Area twelve years ago, I came in through a different door. I did my Master’s in Computer Science at San Francisco State University, and I spent the next several years moving between two worlds. The Stanford students I met were smart, kind and hardworking. So were the SFSU students. The difference wasn’t talent. It was everything around the talent.</p><p>Many of my SFSU classmates worked grocery store shifts to support their families and had to leave class for them. At Stanford career fairs, companies pay thousands of dollars for a booth. At SFSU, we leaned on alumni to bring employers in at all. A Stanford CS student picks between competing offers and summer internships abroad; an SFSU student is lucky to land one internship. The students are not the variable. The investment is.</p><p>Two things changed my trajectory. The first was <a href="https://metro.sfsu.edu/">SFSU’s Metro College Success Program</a>, where I analyzed mentoring data to understand why students drop out. The answers looked nothing like the Stanford dropout story of founding a startup. They were about rent, family, exhaustion. The second was <a href="https://www.codepath.org/about">CodePath</a>, a nonprofit founded by Stanford alum <a href="https://www.linkedin.com/in/timothy1ee/">Tim Lee</a> to open tech careers to CS students from any school. Through a CodePath program, I earned a security internship at Facebook and that internship started my career.</p><p>I didn’t need a different brain. I needed someone to believe in me before I had the credentials to justify it.</p><h3>What the access is for</h3><p>That’s the pattern Chuck and I keep seeing, and it’s why we started the <a href="https://www.zhoueesleyfoundation.com/">Zhou &amp; Eesley Family Foundation</a> in 2021, the year COVID made the gap impossible to ignore, when some students attended Zoom class from a family yacht and others from a shared bedroom. Over the past five years we’ve worked with schools and organizations in Vietnam, Tanzania, Taiwan, Malaysia and many others. <strong>Education compounds fastest where the baseline investment is lowest.</strong></p><p>The most encouraging part has been how many Stanford alums have joined us. They’re the answer to Baker’s question: what the access is for.</p><p><a href="https://www.linkedin.com/in/james-okaro-juma/"><strong>James Juma</strong></a><strong> </strong>spent years teaching at <a href="https://lohada.org/">LOHADA Children’s Home</a> in Tanzania, where he adapted the entrepreneurship curriculum he’d learned at Stanford so his students could use it to solve problems in their own community. Beyond the classroom, he’s been steering LOHADA toward food and financial sustainability, running international fundraising aimed at investing in the home’s own food production and social enterprises, so the kids aren’t dependent on outside aid forever. James led our <a href="https://www.zhoueesleyfoundation.com/programs/tanzania/">teaching trip in Tanzania in 2023</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*N_q4VkKhOkG0alsHdrBTPw.jpeg" /><figcaption>James Juma teaching alongside Chuck at LOHADA Children’s Home, Tanzania, 2023.</figcaption></figure><p><a href="https://www.linkedin.com/in/abisolacorettakusimo/"><strong>Abisola Kusimo</strong></a> flew to Tanzania to join that same 2023 trip right after her Stanford commencement. She taught our future-careers class still wearing her doctoral hat from the mechanical engineering PhD she’d received days earlier. The hat wasn’t a gimmick. As she put it: you cannot be who you cannot see. For a room full of girls in Tanzania, a Black woman with a Stanford PhD in engineering wasn’t an abstraction anymore.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tEZeBEPI4WVbMGsCrBxABQ.jpeg" /><figcaption>Abisola Kusimo teaching our future-careers class at LOHADA in 2023, still wearing her Stanford doctoral cap. “You cannot be who you cannot see.”</figcaption></figure><p><a href="https://www.linkedin.com/in/baotphan/">Bao Phan</a> was born in Vietnam and left as a child. When she joined our <a href="https://www.zhoueesleyfoundation.com/programs/vietnam/">2024 teaching trip</a>, it was her first time back. That trip and her drive to give back to where she came from opened the doors to our current partnerships with <a href="https://www.fuvusa.org/">Fulbright University Vietnam</a> and <a href="https://www.ueh.edu.vn/en/">UEH College of Business</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6IYQ0vNskt3_R4_5CHuOvg.jpeg" /><figcaption>Bao Phan teaching AI-driven marketing workflow at UEH College of Business.</figcaption></figure><p>This is the Stanford I want to talk about. <strong>Elite education is real, and the access it provides is real. The question is what you do with it. Ruling the world is one option. Serving it is another. And from where I’m standing, it’s the better return.</strong></p><p>If you’d like to join us, you can find our work at <a href="https://www.zhoueesleyfoundation.com/contact/">zhoueesleyfoundation.com</a>.</p><p><em>Originally published at </em><a href="https://open.substack.com/pub/zhoulijie/p/how-to-rule-the-world?r=3g0q3&amp;utm_campaign=post-expanded-share&amp;utm_medium=web"><em>https://zhoulijie.substack.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b1b3b0232c91" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My first AI RAG with Milvus and Twelve-Lab]]></title>
            <link>https://lijie-zhou.medium.com/my-first-ai-rag-with-milvus-and-twelve-lab-6287b79ddc10?source=rss-e44b1724aa04------2</link>
            <guid isPermaLink="false">https://medium.com/p/6287b79ddc10</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[women-in-ai]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[rags]]></category>
            <category><![CDATA[milvus]]></category>
            <dc:creator><![CDATA[Lijie Zhou]]></dc:creator>
            <pubDate>Wed, 29 Jan 2025 06:14:17 GMT</pubDate>
            <atom:updated>2025-01-29T06:15:35.497Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*p7F-cjSz7hwuJPPSjDnXXA.png" /></figure><p>This past weekend, I participated “𝗪𝗼𝗺𝗲𝗻 𝗶𝗻 𝗔𝗜 𝗥𝗔𝗚 𝗛𝗮𝗰𝗸𝗮𝘁𝗵𝗼𝗻” organized by <a href="https://www.linkedin.com/company/women-and-ai/posts/?feedView=all">Women and AI</a>. In less than 5 hours, a group of amazing women came up ideas, learned tools sponsored by sponsors from top-notch AI companies and solved problems we care deeply. This is the first time I tried to build an AI RAG, and I learned a lot from all the passionate female participants, experienced mentors, enthusiastic organizers.</p><p>My group was comprised with me and <a href="https://www.linkedin.com/in/silvana-moiceanu/">three </a><a href="https://www.linkedin.com/in/jenniferbtran2002/">other</a> <a href="https://www.linkedin.com/in/irisyu2000/">talented </a>women who want to help SMB. We all know these medium/small business owners usually have a ton to manage in their day to day and they depend heavily on social media to scale their marketing. However, they usually don’t have the budget to hire a social media manager.<strong> Therefore, we came up with the idea to help them using AI to automate their social media marketing.</strong></p><p>In order to achieve this goal, we decided to use <a href="https://www.twelvelabs.io/">Twelve-Labs’ API </a>to analyze video contents shared on the Internet about their products, process the long video into short clips and find the most popular clips to post on the SMB’s social media. We also chose to store the embedding generated by the API in the <a href="https://milvus.io/">Milvus Vector database</a> so that we can continue to generate short clips based on different marketing campaigns the SMB want to run.</p><p><strong>Step 1: Import libraries and connect with local mulvis database</strong></p><pre>from pymilvus import MilvusClient<br>from twelvelabs import TwelveLabs<br>from twelvelabs.models.embed import EmbeddingsTask<br>import os<br><br>milvus_client = MilvusClient(&quot;milvus_twelvelabs_demo.db&quot;)<br><br>print(&quot;Successfully connected to Milvus&quot;)<br><br>collection_name = &quot;twelvelabs_demo_collection&quot;<br><br>if milvus_client.has_collection(collection_name=collection_name):<br>    milvus_client.drop_collection(collection_name=collection_name)<br><br>milvus_client.create_collection(<br>    collection_name=collection_name,<br>    dimension=1024  # The dimension of the Twelve Labs embeddings<br>)<br><br>print(f&quot;Collection &#39;{collection_name}&#39; created successfully&quot;)</pre><pre>Successfully connected to Milvus<br>Collection &#39;twelvelabs_demo_collection&#39; created successfully</pre><p><strong>Step 2: Initiate Twelve-lab Client using its API key</strong></p><pre>TWELVE_LABS_API_KEY = os.getenv(&#39;TWELVE_LABS_API_KEY&#39;)<br>twelvelabs_client = TwelveLabs(api_key=&quot;YOUR_TWELVE_LAB_API_KEY&quot;)</pre><p><strong>Step 3: Generate embedding using Twelve-lab API</strong></p><pre>def generate_embedding(video_url):<br>    &quot;&quot;&quot;<br>    Generate embeddings for a given video URL using the Twelve Labs API.<br><br>    This function creates an embedding task for the specified video URL using<br>    the Marengo-retrieval-2.6 engine. It monitors the task progress and waits<br>    for completion. Once done, it retrieves the task result and extracts the<br>    embeddings along with their associated metadata.<br><br>    Args:<br>        video_url (str): The URL of the video to generate embeddings for.<br><br>    Returns:<br>        tuple: A tuple containing two elements:<br>            1. list: A list of dictionaries, where each dictionary contains:<br>                - &#39;embedding&#39;: The embedding vector as a list of floats.<br>                - &#39;start_offset_sec&#39;: The start time of the segment in seconds.<br>                - &#39;end_offset_sec&#39;: The end time of the segment in seconds.<br>                - &#39;embedding_scope&#39;: The scope of the embedding (e.g., &#39;shot&#39;, &#39;scene&#39;).<br>            2. EmbeddingsTaskResult: The complete task result object from Twelve Labs API.<br><br>    Raises:<br>        Any exceptions raised by the Twelve Labs API during task creation,<br>        execution, or retrieval.<br>    &quot;&quot;&quot;<br><br>    # Create an embedding task<br>    video_task = twelvelabs_client.embed.task.create(<br>        model_name=&quot;Marengo-retrieval-2.7&quot;,<br>        video_url=video_url<br>    )<br>    print(f&quot;Created task: id={video_task.id} model_name={video_task.model_name} status={video_task.status}&quot;)<br><br>    # Define a callback function to monitor task progress<br>    def on_task_update(video_task: EmbeddingsTask):<br>        print(f&quot;  Status={video_task.status}&quot;)<br><br>    # Wait for the task to complete<br>    status = video_task.wait_for_done(<br>        sleep_interval=2,<br>        callback=on_task_update<br>    )<br>    print(f&quot;Embedding done: {status}&quot;)<br><br>    # Retrieve segmented video embeddings<br>    # task_result = twelvelabs_client.embed.task.retrieve(task.id)<br>    video_task = video_task.retrieve()<br><br>    video_segments = video_task.video_embedding.segments<br>    print(f&quot;Retrieved {len(video_segments)} video segments&quot;)    <br><br>    # Extract and return the embeddings<br>    video_embeddings = []<br>    for segment in video_segments:<br>        video_embeddings.append({<br>            &#39;embedding&#39;: segment.embeddings_float,<br>            &#39;start_time&#39;: segment.start_offset_sec,<br>            &#39;end_time&#39;: segment.end_offset_sec,<br>            &#39;embedding_scope&#39;: segment.embedding_scope<br>        })<br>    <br>    return video_embeddings, video_segments</pre><p>It takes some time to break a long video into small clips and generate the vector.</p><pre>Processing video URL: http://commondatastorage.googleapis.com/gtv-videos-bucket/sample/BigBuckBunny.mp4<br>Created task: id=679996df726e6f84ec51414d model_name=Marengo-retrieval-2.7 status=processing<br>  Status=processing<br>  Status=processing<br>  Status=processing<br>  Status=processing<br>  Status=processing<br>  Status=processing<br>  Status=processing<br>  Status=processing<br>  Status=processing<br>  Status=processing<br>  Status=processing<br>  Status=processing<br>Embedding done: ready<br>Retrieved 100 video segments</pre><p>After testing this function with one URL, we can make another function to process a list of videos as the following:</p><pre>def generate_embeddings_for_videos(video_urls):<br>    &quot;&quot;&quot;<br>    Generate embeddings for multiple video URLs using the Twelve Labs API.<br><br>    Args:<br>        video_urls (list[str]): A list of video URLs.<br><br>    Returns:<br>        tuple of two lists:<br>            1. A list where each element corresponds to the embeddings<br>               for one of the provided video URLs.<br>            2. A list where each element corresponds to the video segments<br>               for one of the provided video URLs.<br>    &quot;&quot;&quot;<br>    all_video_embeddings = []<br>    all_video_segments = []<br><br>    for url in video_urls:<br>        print(f&quot;Processing video URL: {url}&quot;)<br>        video_embeddings, video_segments = generate_embedding(url)<br>        all_video_embeddings.extend(video_embeddings)<br>        all_video_segments.extend(video_segments)<br><br>    return all_video_embeddings, all_video_segments</pre><p><strong>Step 4: (Optional) Save the embedding in a pickle before inserting into Milvus</strong></p><pre>import pickle<br>with open(&#39;embeddings.pkl&#39;, &#39;wb&#39;) as f:<br>    pickle.dump({&#39;video_embeddings&#39;:all_video_embeddings, &#39;video_segments&#39;:all_video_segments}, f)</pre><pre>with open(&#39;embeddings.pkl&#39;, &#39;rb&#39;) as f:<br>    data = pickle.load(f)<br><br>video_embeddings = data[&#39;video_embeddings&#39;]<br>video_segments = data[&#39;video_segments&#39;]</pre><p>I found this is a very useful trick when debugging. I don’t have to keep calling the Twelve-lab API nor waiting for the video to be processed.</p><p>You can also print out the embedding and decide whether they are indeed what you want to store in the vector database.</p><pre>print(f&quot;Generated {len(video_embeddings)} embeddings for the video&quot;)<br>for i, emb in enumerate(video_embeddings):<br>    print(f&quot;Embedding {i+1}:&quot;)<br>    print(f&quot;  Scope: {emb}&quot;)<br>    print(f&quot;  Scope: {emb[&#39;embedding_scope&#39;]}&quot;)<br>    print(f&quot;  Time range: {emb[&#39;start_time&#39;]} - {emb[&#39;end_time&#39;]} seconds&quot;)<br>    print(f&quot;  Embedding vector (first 5 values): {emb[&#39;embedding&#39;][:5]}&quot;)<br>    print()</pre><pre>Generated 248 embeddings for the video<br>Embedding 1:<br>  Scope: {&#39;embedding&#39;: [0.0068783234, -0.00729163, 0.032936763, 0.0147475805, 0.0004509193, -0.058732983, 0.010216852, -0.0107900575, 0.013492671, -0.037403982, -0.021999722, -0.03039865, -0.0019278503, 0.01788534, 0.012924736, 0.011753899, -0.024063474, -0.09012038, 0.021896683, 0.006487544, 0.010387876, -0.02762893, 0.06674046, 0.04259518, 0.012998299, -0.01387637, 0.007397698, -0.010834219, -0.02403517, -0.031014148, -0.011390642, 0.0076335417, -0.043314736, 0.026069751, -0.023581678, -0.0037833375, 0.014594813, -0.0064795073, -0.01648418, -0.026304655, -0.011824551, 0.012401233, -0.07024261, -0.010345327, 0.033803698, 0.03137717, 0.020379616, -0.008056862, -0.03816837, -0.016470522, -0.023214314, 0.051605448, 0.012382858, 0.024991678, 0.025948202, -0.056700896, -0.05002672, 0.005621545, 0.003799754, 0.017514924, 0.009348112, -0.014273964, -0.016253011, -0.008670815, 0.021687822, -0.005776424, -0.00073867704, 0.036776777, -0.006435954, 0.014411234, 0.017932959, 0.018801862, 0.01262726, -0.0006321856, 0.030710502, -0.020881118, 0.045263916, 0.009066172, 0.015938517, 0.03683724, -0.02392934, 0.0014410899, -0.018138453, -0.0027101038, 0.03221234, -0.010062318, -0.05035674, 0.0065278504, -0.0095600365, -0.017131425, -0.004740929, -0.001152888, -0.014686445, 0.004312696, 0.033370294, -0.046755593, 0.041128732, 0.0011049962, -0.0326549, -0.010974298, 0.009355865, -0.026824996, 0.0059923884, -0.007159826, 0.02765087, -0.011514687, 0.0029979842, 0.02120983, -0.04027363, 0.038482137, -0.01641928, 0.008217713, -0.020805424, -0.024270693, -0.007831477, -0.0027779758, 0.016611941, -0.015944552, -0.018733263, -0.00852722, 0.009824787, 0.10373483, -0.034193404, -0.05750813, -0.0076109054, -0.0046203965, 0.04162599, -0.00020951532, 0.020191642, -0.0025768424, -0.011773944, -0.0058374093, 0.030586433, 0.013864975, 0.03457791, -0.031917118, -0.029103419, 0.0041772886, -0.0002820209, -0.03795695, 0.011982762, -0.009750477, 0.009444764, 0.0053242138, -0.018103287, -0.0014880485, -0.0016208793, -0.011276391, -0.011566789, -0.009240858, -0.058047164, -0.0015389281, -0.07731879, 0.042317398, -0.018264148, 0.02627899, -0.030914245, 0.032359302, 0.0115260575, -0.0069430806, 0.009181425, 0.009486549, -0.007950507, 0.014979703, 0.0142746745, 0.007965513, 0.038950227, -0.14187975, -0.020468578, -0.028338984, 0.040064123, -0.025516056, -0.06857561, 0.036839914, 0.03944069, -0.022443691, -0.028300164, -0.018598618, -0.01300676, -0.030894853, -0.017789967, 0.0038162016, 0.016857753, -0.032929346, -0.036539637, -0.0059751403, -0.013274721, -0.007999669, -0.016166275, -0.018527169, -0.024972428, 0.023004191, -0.04161817, -0.018189315, 0.02648446, 0.06481213, 0.006909985, -0.009575803, 0.015778968, -0.056017257, 0.02846466, -0.0017617943, 0.0039658663, 0.0040127416, 0.01011397, -0.007591392, -0.03371089, -0.020009829, -0.018150965, -0.07622767, -0.022820394, 0.019242376, 0.029030535, -0.0037084634, -0.0052097123, -0.033088524, -0.0013969197, 0.011983584, 0.020200059, -0.0139105385, 0.023957409, 0.01630936, 0.012070444, 0.024898294, 0.0267731, -0.0072839046, -0.036392916, 0.009463345, 0.012967374, -0.009006891, -0.051524155, 0.023945177, 0.018197874, 0.009634776, -0.0060075605, 0.0027702616, 0.023907999, -0.034060962, 0.027866267, 0.01902927, -0.00663021, 0.035967648, 0.0054591126, -0.012791728, 0.03913613, -0.01310292, 0.017438207, -0.039148726, -0.033387296, 0.005630618, -0.028434727, 0.027561612, 0.014691756, 0.07976848, -0.03223561, 0.0007481973, 0.04210834, 0.009957227, 0.008004614, 0.002332506, 0.009767064, 0.03514584, -0.019225692, -0.011511926, 0.010415811, 0.0007344091, -0.022173109, -0.027303904, -0.048365302, 0.009811818, -0.02198606, -0.03649355, -0.0045325044, 0.047147807, -0.014227006, 0.02017479, -0.054162286, 0.009273572, -0.0069489134, -0.042271137, -0.042640712, 0.019760981, 0.01675835, 0.032542817, -0.0066696703, 0.031744607, 0.0082657635, 0.03653774, 0.038985252, 0.041337114, -0.008488836, -0.0024951724, 0.0038448551, -0.05569678, 0.007930949, -0.017632779, 0.03113436, -0.0015900102, 0.016602702, 0.03158648, 0.028589448, -0.016444702, -0.023192272, -0.035195976, -0.0076671825, 0.0027764214, 0.03575938, -0.019178644, 0.0020819793, -0.008073123, -0.003904652, -0.08572666, 0.029566716, -0.0116582895, 0.012402559, -0.049744148, -0.024174564, -0.004230353, 0.030347487, -0.023778526, -0.02489188, -0.008305566, -0.0017777489, 0.036159556, 0.0037250384, -0.0150255, -0.022011433, 0.041830394, 0.021162769, -0.024522621, 0.056461897, 0.020509608, -0.026023943, 0.05704436, 0.007657682, 0.020544255, -0.038667798, -0.015546326, -0.03882337, 0.043772332, -0.008450909, -0.002395437, -0.0032268916, 0.042917456, 0.0028435257, 0.03455453, 0.019187514, -0.031819936, 0.047565334, 0.010991625, 0.01090406, -0.027776588, -0.025718905, 0.0074462374, -0.004514669, -0.013848633, 0.02666192, 0.0085935425, -0.010388193, -0.031570036, 0.011940553, -0.023316042, 0.007985697, -0.009717924, -0.032650687, -0.015240673, 0.006252738, 0.01639453, -0.011811621, 0.014626912, 0.008785405, 0.0031616385, 0.0019156403, 0.026496114, 0.023236208, 0.024150465, -0.0035217493, -0.0082648005, -0.0027253532, 0.0072417166, -0.047684617, 0.028748259, 0.003027111, -0.022774346, 0.015822295, -0.005282118, -0.021030404, 0.012538949, 0.03133843, 0.0023226822, 0.0022601306, 0.02122433, -0.032834265, -0.034208894, -0.04677248, 0.045770034, 0.0034248435, 0.05710468, -0.051385716, -0.006663521, -0.030790245, 0.047688797, 0.016293691, -0.02445442, -0.0001873563, -0.013941138, 0.012366109, 0.11104746, -0.010519957, 0.025310112, 0.0078193275, 0.031424575, -0.044816557, -0.0019667696, -0.02397321, 0.04276926, -0.017788978, -0.00660508, -0.01770767, -0.020226305, 0.014092322, -0.04236608, -0.006318598, -0.017398776, 0.031128887, 0.042356927, -0.011390976, 0.026528185, 0.0024001093, 0.0045336774, 0.0031395424, -0.06361563, 0.010888834, -0.014075934, 0.0031002604, 0.004764683, -0.015869934, -0.035132155, 0.0128163695, -0.010045604, 0.0036366112, -0.029119384, 0.010648713, 0.009941925, 0.035310056, 0.016759366, -0.1439708, -0.0005380146, 0.046135873, 0.034784123, -0.009871211, 0.022185007, -0.0014119252, -0.032703836, 0.030556154, 0.014461451, 0.018753773, 0.02155308, -0.02236582, 0.020895544, -0.043511555, 0.04225371, 0.014533836, 0.009618576, 0.026846372, 0.020152451, -0.07046795, -0.03666141, 0.00030562255, -0.023430772, -0.0022476863, 0.007895782, -0.007577382, -0.0714504, -0.01354767, 0.037307434, -0.0042871707, 0.012452606, -0.004152858, -0.034020983, -0.00046625422, 0.011293103, -0.044560302, 0.0064419378, -0.04632161, -0.0566079, -0.0312156, -0.018052597, -0.017313259, 0.0023549816, -0.042245153, 0.008310225, 0.0019696637, -0.002121526, 0.02606576, 0.009925641, -0.026181472, 0.007890991, -0.016468776, -0.00036539926, -0.018127602, 0.032954507, 0.025526823, -0.02157748, -0.0041952073, 0.009162711, 0.02940634, 0.10357208, -0.0024835044, -0.04293473, 0.034040205, 0.017996231, -0.008149953, -0.005262708, 0.0122775305, 0.0124413185, -0.03506368, -0.036942933, -0.005312987, -0.030262979, -0.022831239, -0.0247437, 0.044873904, -0.029957822, -0.0025730622, 0.031658486, 0.003925599, -0.053271662, 0.039127037, 0.05991646, -0.00018395309, -0.03011774, 0.008939936, 0.04975966, 0.015902562, -0.0337514, 0.018274201, 0.03678602, -0.0056254575, -0.019258512, 0.0012723661, -0.031829555, 0.0139304, -0.0050548385, 0.012326478, -0.015357915, 0.038541213, -0.0012350978, -0.030668447, -0.0034853932, -0.026633028, -0.008495682, 0.0034360234, 0.038077015, 0.03716065, 0.019170467, -0.0121424915, -0.035390727, -0.04340154, 0.0034386178, -0.018596359, -0.023703314, 0.015216682, -0.029170884, 0.016857164, -0.04846999, -0.019350732, 0.020737777, 0.010268543, 0.027833402, -0.009738148, 0.023663672, 0.018375307, 0.013672182, 0.035967868, -0.026621914, 0.009072047, -0.011291309, -0.019593613, -0.0050118393, 0.001232954, 0.049719576, -0.049359478, -0.009153291, -0.0009921377, 0.027191922, -0.004971644, -0.005347495, -0.021784484, 0.015026974, 0.031328898, -0.06818599, -0.03839889, 0.05000018, -0.002561457, 0.0016570865, -0.006161121, 0.0099220555, -0.009738079, -0.009359508, 0.045425758, 0.0065615354, -0.0072403527, -0.03345994, 0.083168894, 0.02255066, 0.009120352, 0.015185575, -0.020006701, 0.01602376, -0.034417454, -0.36696416, 0.009146621, -0.0037223706, -0.01410639, -0.019152572, -0.02245828, 0.011676281, 0.02819041, -0.027526101, 0.016556587, 0.019962437, 0.008079883, -0.0036769419, -0.010620434, 0.0046598148, 0.053947378, -0.04319025, -0.009818431, -0.022526842, 0.04379876, -0.00988221, -0.01756134, -0.013981344, 0.00043138862, 0.039103564, -0.012385938, 0.03673933, -0.073141016, 0.010517789, 0.027442463, 0.01980769, -0.02375511, -0.02850647, -0.056625858, -0.027542904, -0.023711696, 0.043005962, 0.016118685, -0.00319455, -0.03216963, -0.059748415, -0.00040207442, 0.026238123, 0.0006560515, -0.005858329, -0.041938268, 0.01639737, -0.03163517, 0.0058489735, 0.0078515485, -0.01581838, -0.004465798, -0.004216052, 0.027780123, 0.03204899, -0.011159788, 0.0036594975, -0.048510022, 0.04918587, 0.026628617, 0.034543708, -0.008005352, -0.0011241935, -0.024024593, -0.007353092, -0.018263623, -0.014880418, 0.044086773, -0.007500205, 0.019337451, -0.02837935, -0.03429122, -0.02299825, -0.013846107, -0.03475906, -0.048231505, 0.004291552, 0.009804889, 0.01503584, 0.032066077, -0.0052789175, -0.029120348, 0.065041594, -0.010063236, -0.046960287, 0.03466021, 0.04549871, 0.0042959685, 0.021595178, -0.065630294, -0.027512921, 0.030167883, 0.031111006, -0.04569625, -0.036343217, 0.008531117, 0.015952036, 0.027329003, -0.008429641, 0.032100048, 0.010928727, -0.031277698, -0.00032509636, 0.023837226, -0.00612003, 0.010887802, 0.01794498, 0.0504663, -0.006652021, 0.021194883, -0.01834971, -0.06998811, 0.029119547, 0.008480561, 0.036035627, 0.009220499, 0.0044326154, 0.0011677087, -0.027754717, 0.016578382, -0.01715151, 0.017734673, 0.037889484, 0.023782177, -0.02695164, 0.023280395, -0.020525906, -0.0061298953, -0.032623664, -0.03457432, 0.009564498, 0.0017400886, 0.043605108, 0.014460223, 0.0006737991, 0.012836562, -0.030985877, 0.014578673, -0.017155439, 0.022641493, 0.0106985355, -0.038595032, -0.06655343, -0.012888666, -0.04074178, 0.033100277, 0.0036092966, 0.06446328, 0.028097412, -0.0044544856, -0.02155137, -0.014769593, -0.042888578, 0.0114436345, 0.00612343, 0.0032519381, -0.027427435, -0.03514072, -0.13859929, -0.025048153, 0.033394016, 0.009462309, -0.036709767, 0.028556371, 0.04795794, 0.035540916, -0.049496565, -0.006405567, 0.0030413924, 0.015967362, 0.018966781, -0.012907407, 0.020308798, -0.01032781, -0.010573173, 0.03340878, 0.0037400026, -0.031975143, -0.023609469, 0.0012232695, 0.012158817, 0.01569287, -0.032557283, -0.016794557, 0.00044435635, -0.021173855, -0.015432922, 0.02630522, 0.0005279225, -0.0025921136, -0.015539128, 0.075713634, -0.0027337233, 0.028622676, -0.02356655, 0.018520506, -0.015336219, 0.00960078, 0.0039478266, -0.027863234, -0.012071628, 0.046242293, -0.028982809, 0.013574303, 0.027919203, 0.03210301, 0.047823623, -0.025815012, -0.0032278604, 0.025732758, -0.022937106, 0.009831685, -0.020237489, -0.015523589, 0.014118576, 0.007623117, 0.019190995, 0.009425359, -0.0067071198, 0.06481179, 0.016524622, -0.024244584, 0.056911364, -0.015880961, 0.039218206, 0.00043134534, 0.015957633, -0.010186033, -0.01305188, 0.007632062, -0.03146358, -0.029776895, -0.07687758, -0.014195089, -0.01269619, -0.109493986, 0.014518725, -0.003920082, 0.029761791, 0.028780308, 0.042023815, 0.017625073, -0.027262283, 0.006201132, 0.0013434504, -0.034603447, 0.04516391, -0.034812525, -0.0019955463, -0.011706088, 0.027777651, 0.05312186, 0.03384893, -0.017940093, -0.0771205, -0.017582705, 0.024866663, -0.035245355, -0.0035456396, 0.005352694, 0.02640106, 0.05344151, 0.010181227, 0.0139562525, -0.02177064, -0.015945155, 0.02389725, 0.0018519562, -0.04212108, -0.02762105, -0.012241074, -0.028351039, -0.0054267747, -0.04709828, -0.051129874, 0.034532957, 0.009898679, 0.039978884, 0.0765568, -0.01063408, -0.024362851, 0.028029755, -0.0022016144, -0.01737006, 0.016511118, -0.0057703126, -0.0068922443, 0.030503392, 0.001122857, -0.0009620622, 0.028873384, 0.0006953626, -0.009667843, 0.02817171, -0.023649752, -0.017447026, 0.013218727, 0.04172558, 0.02223032, 0.032463603, -0.020011622, 0.01268178, -0.035604723, 0.030376436, 0.009219191, -0.03439211, 0.025030501, -0.00090045907, -0.040826384, 0.03889264, -0.042424195, 0.02332162, -0.06767599, 0.033809923, 0.0003984815, 0.059741758, -0.014035683, -0.02910125, -0.015128265, -0.0011207975, 0.026216427, 0.0090320455, 0.009699141, -0.0025208262, -0.022278797, -0.046547826, 0.037042085, -0.005513966, 0.026397517, 0.0058617303, -0.011617904, -0.003821055, -0.006527486, -0.043904696, -0.00080996385, 0.02699345, 0.06422825, 0.0077991853, 0.015246814, -0.02129638, 0.057596937, -0.038282506, -0.045133162, 0.02327624, -0.0077782045, 0.0029991404, -0.0049662325, -0.00037101674, -0.018386306, 0.036274984, 0.0011811958, -0.021305962, -0.052296497, 0.022840908, 0.028204726, -0.0003874616, 0.027440667, -0.011601457, -0.028136825, -0.019587373, 0.027002899, -0.050302148, -0.12951034, 0.015779821, -0.0073972195, -0.039541904, 0.0054853787, -0.014319935, -0.033774856, -0.03204703, 0.032221172, -0.0065898853, -0.013174562, 0.0317752, -0.0011760573, 0.026424585, -0.003753385, 0.005472652, -0.0039887596, -0.037580777, 0.0075959335, -0.027274583, 0.009036218, -0.0052845576, -0.0022934363, -0.03871653, 0.0067554777, 0.040892716, -0.028265215, 0.00516633, -0.013281223, -0.03756754, 0.0272351, 0.033563476, -0.023648096, -0.030575361, 0.04350686, -0.015491724, -0.00816973, -0.015462358, -0.020979615, 0.03511203, 0.0012422261, 0.009834583, 0.001139739, 0.017173205, -0.0122829145, 0.036878075, 0.027375592, 0.009843759, -0.012545329, -0.0058124964, -0.013636089], &#39;start_time&#39;: 0.0, &#39;end_time&#39;: 6.0, &#39;embedding_scope&#39;: &#39;clip&#39;}<br>  Scope: clip<br>  Time range: 0.0 - 6.0 seconds<br>  Embedding vector (first 5 values): [0.0068783234, -0.00729163, 0.032936763, 0.0147475805, 0.0004509193]<br><br>Embedding 2:<br>  Scope: {&#39;embedding&#39;: [0.0074522155, -0.008839504, 0.034555014, 0.01722106, -0.010919876, -0.047983438, -0.003969231, 0.015394325, 0.019325182, -0.03302268, 0.0036118526, -0.03237224, -0.012218506, 0.031316873, 0.024598695, 0.0010986571, -0.043801244, -0.0875056, 0.013996466, 0.0002929803, 0.0056979507, -0.035210636, 0.09175682, 0.055135105, -0.0015013883, -0.0027936823, 0.010230771, 0.020050872, 0.0053680036, -0.01654871, -0.0040683947, 0.0036912856, -0.06558114, 0.010166385, -0.010372616, -0.015901484, 0.027263124, 0.00090340513, -0.020741116, -0.03606464, -0.04151167, 0.010574331, -0.044444386, -0.018270578, 0.022846475, 0.0126374625, -0.01605193, 0.0055646966, -0.047296666, 0.0016428804, -0.020698031, 0.034028556, 0.010037179, 0.018126732, 0.034498848, -0.04462811, -0.05053957, 0.0016068524, 0.016463613, 0.019203825, 0.012457982, 0.016695637, 0.015533181, -0.0329491, 0.01952508, -0.021674145, -0.0038345966, 0.023405502, 0.009713134, 0.024041677, 0.032211713, 0.012576292, 0.015988128, -0.017023928, 0.016136654, -0.029923808, 0.0327615, -0.0057500796, 0.01511292, 0.023775356, -0.03124114, 0.015777012, -0.024296083, -0.0221823, 0.025306853, 0.012979236, -0.046464603, -0.006332406, -0.0003593868, -0.038349815, -0.024593312, 0.0023170174, -0.018886244, 0.011246997, 0.028603066, -0.044414733, 0.031320117, 0.0062158387, -0.02696245, 0.015389885, -0.0044665015, -0.023483284, 0.011541991, 0.007013622, 0.029881889, 0.011539056, -0.027291762, 0.035607666, -0.026082454, 0.057297852, -0.024961868, 0.03989055, -0.031078985, -0.022983737, -0.0018090889, 0.012321865, 0.013564377, -0.020751193, -0.010411593, -0.010858802, 0.0134096965, 0.060112998, -0.035103463, -0.04701275, -0.017446078, 0.0010710568, 0.060341157, 0.010150115, 0.019090725, -0.0124133555, -0.016380023, 0.0021087802, 0.008104993, 0.0035663773, 0.044184078, -0.020512225, -0.018875273, 0.003000773, 0.008793117, -0.04732549, 0.0017348255, -0.020439202, 0.0029288498, 0.016255625, -0.024763567, 0.009089848, 0.0052437005, -0.028627027, 0.0050714393, -0.006789734, -0.05497339, -0.008503456, -0.061290164, 0.042472232, -0.015484782, 0.034256827, -0.036693066, 0.01392842, 0.021778222, 0.01473207, 0.015598581, 0.008234391, -0.015803503, 0.005353044, 0.030985624, 0.021291142, 0.030982573, -0.13186565, -0.033725522, -0.040277682, 0.027824402, -0.023747055, -0.058825646, 0.050798398, 0.012941631, -0.033114273, -0.039239477, 0.0007844426, -0.043092333, -0.03740829, -0.015075767, 0.01506626, 0.028269494, -0.032943457, -0.023515321, 0.0007020762, -0.013120824, -0.0062892986, 0.0018166205, -0.009568889, -0.004877579, 0.03286323, -0.048344996, 0.0019486942, 0.022634268, 0.03317508, -0.01992588, -0.030192612, 0.022147069, -0.03224373, 0.016191566, 0.000319139, 0.036480516, 0.0127610145, 0.01653903, -0.026093874, -0.026741553, -0.021060588, -0.019100532, -0.06397619, -0.015823204, -9.500625e-05, 0.03792768, -0.01984515, 0.00959305, -0.050971203, 0.0056934957, 0.021773197, 0.024332585, -0.021953136, 0.0144275855, 0.029886246, 0.026095865, 0.0250103, 0.018261673, -0.0033321916, -0.03262201, 0.009617943, 0.025506664, -0.029454175, -0.054415062, 0.014884838, 0.046369553, 0.016881883, -0.008153895, 0.020554058, 0.012338968, -0.016325356, 0.043196797, -0.007388757, -0.00016400198, 0.05305024, 0.0042506754, -0.044684626, 0.052890655, -0.019529512, -0.017976714, -0.044737395, -0.03261188, 0.019561365, -0.047055382, 0.032494083, 0.023960965, 0.0949476, -0.04875176, -0.02395966, 0.05057814, 0.011857367, 0.033985447, 0.010547104, 0.015338917, 0.00020208677, -0.015378516, -0.015380445, 0.0012680556, -0.0005537721, -0.016271072, -0.03857157, -0.034617215, -0.022016747, -0.02384039, -0.055178117, -0.008344697, 0.054001056, -0.00033679407, -0.011944844, -0.02633296, 0.02382716, 0.008060967, -0.03991643, -0.04616718, 0.003658511, 0.018253265, 0.051839393, -0.007895639, 0.028671525, 0.03122211, 0.038805302, 0.028986463, 0.04190596, 0.00063219474, 0.009065546, -0.010198142, -0.037055697, 0.03898748, -0.02702048, 0.04070129, -0.01608144, -0.007742948, 0.025886437, 0.03193586, 0.0013145296, 0.0057618115, -0.047126785, -0.010962562, 0.011621482, 0.033822604, -0.010802626, 0.00038705603, -0.011450696, -0.013063889, -0.07076639, 0.019701138, -0.0056664934, 0.017525075, -0.029505204, -0.020629888, -0.008371687, 0.039534606, -0.03263551, -0.0033216707, -0.009457929, 0.0011473531, 0.0509124, -0.0048315045, -0.005110509, -0.022008937, 0.036116883, 0.013796326, -0.015202857, 0.042754635, 0.013805734, -0.034749713, 0.06231565, 0.009444741, 0.028202932, -0.02741537, -0.0079521295, -0.025104241, 0.03919504, -0.013540474, 0.0047395425, 0.008202283, 0.04018482, -0.0103747, 0.04308249, 0.011015371, -0.028648192, 0.029191129, 0.012855809, -0.004737785, -0.026896536, -0.041763585, -0.008788402, 0.01324675, -0.030761193, 0.014724701, 0.009202018, -0.008297771, -0.02618187, 0.01786139, -0.048647836, 0.010163986, -0.010693619, -0.023731768, -0.007677954, 0.013141625, 0.021185853, -0.0088818045, 0.028100643, 0.0055919276, -0.012808517, 0.013315591, 0.024857523, 0.014844072, 0.04541211, 0.009987041, -0.02570884, -0.0138027305, 0.008808417, -0.04329894, 0.023254452, -0.011059145, -0.0114806555, -0.00048643403, -0.017681368, -0.022879975, 0.030874059, 0.047962096, -0.017192444, -0.0039520864, 0.00044442792, -0.033431478, -0.032738198, -0.0277785, 0.04580049, 0.0076631657, 0.039375797, -0.050045773, -0.0071065095, -0.028954815, 0.03491413, 0.008754293, -0.018409047, 0.001350733, 0.00449625, 0.012154206, 0.11494169, -0.017192023, 0.047535278, 0.021609517, 0.014511419, -0.029023308, 0.008903684, -0.023523657, 0.032582138, -0.028731918, 0.025772408, -0.00823772, -0.03336441, 0.0153388735, -0.038759615, -0.0043524266, -0.017254744, 0.026294226, 0.047559313, 0.00046868695, 0.025153868, -0.0013562788, 0.019829318, -0.0068216063, -0.07789945, -0.0095998105, -0.036796525, 0.00082318095, 0.013771439, 0.014961619, -0.01590642, 0.0122878, 0.002124247, 0.0016801918, -0.021132741, 0.0010967798, 0.016331447, 0.039687093, 0.016383503, -0.11057145, 0.020452797, 0.042369287, 0.024737539, -0.01870456, 0.024647364, 0.0080966465, -0.020726392, 0.016426273, 0.026055466, -0.0019631602, 0.008453003, -0.010453186, 0.025560046, -0.03755639, 0.018824542, 0.02759534, 0.010679917, 0.020819431, 0.036461547, -0.10842315, -0.054774504, -0.025548685, -0.022423698, 0.019568136, 0.017678395, 0.013554157, -0.061949905, 0.010845249, 0.035112802, 0.026534267, 0.0032890306, -0.002455338, -0.047345523, -0.023355698, 0.021318082, -0.036779266, 0.011641729, -0.03038354, -0.055115458, -0.050654724, -0.03585867, 0.002014872, -0.030505488, -0.027747488, 0.025020111, -0.035633802, 0.0005055557, 0.029098578, 0.0049930927, -0.033487264, 0.023203656, -0.010102894, 0.0066570654, -0.020813286, 0.039696585, 0.014227219, -0.018826427, -0.0024220855, -0.009435663, 0.038692884, 0.100536436, -0.007583739, -0.034818377, 0.06433559, 0.035035636, -0.012604207, -0.0018190952, 0.021554308, 0.0052943095, -0.029224966, -0.030952588, -0.01026406, -0.01749431, -0.03695405, -0.006958224, 0.028516317, -0.03498238, -0.008862354, 0.056704707, 0.0006278681, -0.04560109, 0.040426925, 0.0633897, -0.027978836, -0.029681958, 0.00930991, 0.04354167, 0.016982352, -0.019670483, 0.012758063, 0.035953827, -0.016581377, -0.02143939, -0.0041154125, -0.047105007, 0.011151526, -0.020785976, 0.011184879, -0.012642138, 0.03415584, 0.002364469, -0.017981695, -0.01097375, -0.017823424, -0.024427457, 0.0032048188, 0.04936191, 0.005945351, -0.0013298568, -0.011693887, -0.028543217, -0.0016265583, 0.017211163, -0.022470642, -0.00038615795, 0.001618149, -0.0138190305, 0.029101966, -0.018460771, -0.017390553, 0.014493804, 0.025946941, 0.023398394, 0.006102012, 0.012728348, -0.013428885, 0.02751905, 0.026210878, -0.0068143583, -0.0012630372, 0.013004682, -0.010651298, 0.012670527, -0.014146632, 0.007890834, -0.06991293, -0.031112997, 0.023549126, 0.01745639, -0.0029845478, 0.0158186, -0.006240649, 0.008620537, -0.018124687, -0.07808634, -0.04172951, 0.0435348, -0.0031082518, -0.015723938, -0.01324094, 0.0046213963, -0.034449983, 0.0093339635, 0.024973778, 0.02173823, -0.017753858, -0.028194783, 0.06341544, -0.0007419612, 0.0016624198, 0.01667961, -0.026314141, 0.01490785, -0.02541232, -0.35052314, -0.0005645358, 0.014064282, -0.008224829, -0.022913437, -0.016075676, 0.0033512325, 0.02160306, -0.02248773, 0.043388035, 0.026400229, 0.0205701, 0.018209122, -0.018017976, -0.024040326, 0.040404268, -0.03200456, -0.009984958, -0.036938567, 0.04777178, 0.0010094708, 0.008602074, -0.026016926, -0.016346652, 0.06874132, -0.01080376, 0.037091758, -0.059469264, 0.022988183, 0.020391444, 0.0023783625, -0.023660664, -0.0144932605, -0.0567562, -0.029794462, -0.024240684, 0.028095672, 0.01867404, -0.0049250866, -0.023101125, -0.032652695, 0.007335222, 0.03364143, -0.0056959954, -0.017144904, -0.04666335, -0.0034366145, 0.0058580874, -0.020921385, 0.0017927405, -0.02440825, 0.007750243, -0.008410392, 0.03329953, 0.039600227, -0.0072443523, 0.0037258347, -0.046972927, 0.04070332, 0.027918449, 0.026739608, 0.009036375, -0.017018972, -0.0316889, -0.0020619538, 0.008970702, -0.010148704, 0.02719024, 0.009352575, 0.0055081723, -0.04724948, -0.04471406, -0.02101042, 0.0042553297, -0.012301238, -0.054271337, 0.020664264, 0.013281572, 0.0054261493, 0.021246096, -0.0011443668, -0.011233788, 0.042090777, 0.0024940164, -0.029581564, 0.02093266, 0.0054173176, -0.008559335, 0.032998085, -0.037799474, -0.028654516, 0.010908923, 0.02790449, -0.020106783, -0.012693741, -0.0050824704, 0.019289773, 0.041426525, -0.011617622, 0.0024756708, 0.0063932207, -0.020814313, -0.0038606112, -0.0005396486, -0.0022170981, 0.029194232, 0.0006077018, 0.07310693, 0.0038024148, 0.029789021, -0.04118184, -0.044093337, 0.033490866, -0.0069081904, 0.025653949, 0.013324008, -0.011750565, 0.013519148, -0.031934, -0.004722553, -0.033347405, 0.015585269, 0.03592709, 0.051081844, -0.022580132, 0.03957452, -0.0025742932, -0.025316693, -0.036999833, -0.024936303, 0.00992483, -0.0006321027, 0.06854505, 0.0113024535, 0.01715792, 0.026546637, -0.020897372, 0.012042098, -0.009350388, 0.013876041, -0.004933547, -0.04400092, -0.061475195, -0.02672715, -0.041773114, 0.019261293, -0.00039507682, 0.03895422, 0.01399835, -0.030617034, -0.025432453, -0.001633505, -0.054495133, 0.008656317, 0.014741583, 0.0207017, -0.023507524, -0.040832043, -0.1507923, -0.04715909, 0.035476297, -0.020142764, -0.03515056, 0.021705437, 0.050200153, 0.019293945, -0.041250207, -0.0057212994, 0.009280262, -0.0015125384, 0.02600183, -0.0149057545, 0.010107692, -0.0020923102, -0.021224324, 0.0034150244, -0.0034757287, -0.030974455, -0.023892695, 0.008227798, 0.012985486, -0.019372743, -0.02252764, -0.004561428, 0.021652967, -0.007894514, -0.007763459, 0.02068022, 0.016155962, -0.01407441, -0.012294118, 0.105960235, -0.012431311, 0.0013390266, -0.056968834, 0.02942808, 0.0051433314, 0.03557072, -0.0057072258, -0.022663886, 0.007404118, 0.04391349, -0.01558437, -0.010228479, 0.026943553, 0.028827323, 0.01843621, -0.0054670773, 0.00601881, 0.005827703, -0.024530584, 0.0018050818, -0.025024364, -0.022613777, 0.02725525, -0.015954638, 0.0117042605, 0.009248966, 0.009560541, 0.016398104, 0.02966306, -0.025507055, 0.06267417, -0.018214216, 0.02984473, -0.009826166, 0.01323789, -0.006801949, -0.0013921192, 0.018284563, -0.0045400974, -0.027067075, -0.070493646, -0.029132422, -0.009164447, -0.092897445, 0.016450109, -0.010583276, 0.023916852, 0.016976863, 0.023459125, 0.026161188, -0.025444625, 0.015295463, -0.014682103, -0.042867552, 0.04336732, -0.028611999, 0.020691326, -0.01541647, 0.048785955, 0.049717303, 0.075828955, -0.03158335, -0.085921, -0.01623374, 0.016249549, -0.03680656, -0.02035232, -0.018181523, 0.028948594, 0.05148784, 0.011370629, 0.045858853, -0.010332891, -0.027784668, 0.010140914, 0.0046650358, -0.03231824, -0.020245485, -0.01593481, -0.007752056, -0.021585377, -0.062865995, -0.057640355, 0.022673344, 0.0007596357, 0.031980623, 0.083702765, -0.0026883192, -0.0091733225, 0.005931313, -0.009190126, -0.00517812, 0.011174043, 0.02562492, 0.014326487, 0.054513097, 0.019465396, 0.0069446885, 0.028103387, 0.0146203665, -0.014074718, 0.0102658365, -0.0329737, -0.027624471, 0.021106796, 0.035920747, 0.02612967, 0.02563972, -0.008541487, 0.020364545, -0.038180467, 0.015401067, 0.0046842345, -0.041359566, -0.0059593706, -0.0076743523, -0.05356275, 0.050124068, -0.03284442, 0.02369669, -0.047801007, 0.03757269, 0.0149502475, 0.029371895, -0.0053442917, -0.057796013, -0.0307359, 0.0186806, 0.0242037, 0.014608835, 0.0052301353, 0.0047475086, -0.03360789, -0.046996195, 0.03944566, -0.013729676, 0.017585816, -0.008653795, -0.0026050278, -0.0057531395, -0.017949287, -0.06877698, -0.0026391724, 0.01967899, 0.06683062, 0.0015945351, 0.003776177, -0.0038557963, 0.04219957, -0.02758097, -0.023152204, 0.009012003, -0.014023754, -0.004570685, -0.002914612, -0.010077189, 0.00078256376, 0.018941144, 0.013118156, -0.0107614845, -0.03160061, 0.02257434, 0.035587814, 0.0028316742, 0.029223422, -0.0008015224, -0.024198629, -0.011273547, 0.026623368, -0.0371461, -0.12988491, 0.017721156, 0.00028679345, -0.04606276, 0.011933843, -0.010357491, -0.043378305, -0.021077305, 0.03520118, 0.0016266997, -0.011232551, 0.028545283, -0.004882632, 0.035530806, -0.008297061, -0.01740012, -0.010268771, -0.032125086, 0.0013169772, -0.018151386, 0.022305503, -0.0019151436, -0.00057913107, -0.05289391, 0.0023381603, 0.03334819, -0.031910345, 0.01605221, 0.021643348, -0.03206316, 0.014681072, 0.022557149, -0.02026251, -0.020424344, 0.014868684, -0.017324904, 0.0054650437, -0.0064178067, 0.0025500422, 0.027708547, -0.02259343, -0.0013550959, -0.021412991, 0.008187213, -0.01229108, 0.02500223, 0.0349292, 0.0049082176, -0.03425189, 0.0061243577, 0.006645685], &#39;start_time&#39;: 6.0, &#39;end_time&#39;: 12.0, &#39;embedding_scope&#39;: &#39;clip&#39;}<br>  Scope: clip<br>  Time range: 6.0 - 12.0 seconds<br>  Embedding vector (first 5 values): [0.0074522155, -0.008839504, 0.034555014, 0.01722106, -0.010919876]<br><br>Embedding 3:<br>  Scope: {&#39;embedding&#39;: [0.022766022, -0.0035799714, 0.019664006, -0.023654077, -0.003662397, -0.005644936, 0.016822277, -3.1653556e-06, 0.00096840377, -0.01660698, 0.035881896, -0.016404742, 0.018058447, 0.011470449, 0.03784944, 0.0056803366, -0.04804558, -0.09573162, 0.005175711, -0.011490836, 0.007233115, -0.020842714, 0.06336342, 0.07920366, 0.0014709332, 0.0134915905, 0.021113815, 0.046989065, 0.04389127, 0.006158011, -0.009976404, 0.00043767973, -0.03882974, 0.016562998, 0.0016153896, -0.011449563, 0.016521443, 0.012509324, -0.02973595, -0.034327567, -0.04099948, 0.016562255, 0.0067015323, -0.026588298, 0.042026445, 0.03530079, -0.035029765, 0.028272832, -0.038650114, 0.02630575, 0.011540845, 0.0020891232, -0.00055229216, 0.055891324, 0.05398184, -0.048230745, -0.021194562, 0.0018502595, 0.007081245, 0.007239543, 0.02400551, 0.026324434, 0.009827246, -0.02530434, 0.0017573369, -0.04127424, 0.007600254, 0.022138173, 0.020989154, 0.05510985, 0.03887861, 0.006920554, 0.010027499, 0.003880198, -0.010817719, -0.045119982, 0.0010119729, 0.0052231844, 0.01725519, 0.0056554596, -0.03620438, -0.002044667, -0.028834924, -0.029801358, -0.014170937, 0.044110566, -0.01893171, -0.0117380265, 0.019670812, -0.04155591, -0.037319683, 0.019337133, -0.0008892915, 0.0011960564, 0.0046740337, -0.053946603, 0.013664724, -0.0019050891, 0.011882006, 0.02431109, 0.0010999193, -0.0043204543, 0.006425615, 0.013913696, 0.010713306, 0.019762151, -0.043771774, 0.006182329, -0.023228092, 0.05833724, 0.004388011, 0.04440492, -0.024188884, -0.043018337, 0.011335503, 0.024822393, 0.018951127, -0.005726927, 0.022509271, -0.01119711, 0.019091316, 0.013284507, -0.0003896782, -0.008828318, -0.04824572, 0.024676291, 0.049399856, 0.007734175, 0.012592344, -0.002084933, -0.052080598, 0.0020062278, 0.016219616, -0.019857178, 0.046692114, 0.010649313, -0.033009488, -0.0007686981, 0.007390102, -0.043722086, -0.0065194624, -0.018470408, 0.0090391, -0.0019626408, -0.030221304, -0.025794713, -0.018214744, -0.010658307, 0.016857028, -0.00466102, -0.0130866775, -0.01467913, -0.041729324, 0.029555643, -0.00090468285, 0.003571088, -0.019921387, -0.004339974, 0.025684344, 0.031013127, 0.0016328557, -0.011420829, -0.017145082, -0.011945379, 0.008222548, 0.032147884, 0.011948834, -0.10673283, -0.011959723, -0.0028394647, 0.0364827, -0.02833538, -0.044244517, 0.015603049, -0.0035074342, -0.022148017, -0.05395577, 0.0030495722, -0.036648292, -0.039928462, -0.0047855596, 0.008438757, 0.016619967, -0.03080492, -0.025995195, -0.028545892, -0.019184032, -0.013782463, 0.013426585, 0.0076742717, 0.027870033, 0.046823043, -0.06724898, 0.004973937, 0.0113658225, 0.029982813, -0.03237467, -0.015540767, 0.012660119, -0.0042782407, 0.020601828, 0.0127076935, 0.037310794, 0.03898782, 0.013098616, -0.0275177, -0.026902584, 0.0010300528, -0.00690599, -0.04704083, -0.0069784853, -0.02670968, 0.027340554, 0.0060593076, -0.0050505586, -0.03225998, -0.017238274, 0.030723076, 0.01883847, -0.007413287, 0.006590858, 0.02751908, 0.0173983, 0.0049086777, 0.010312642, -0.009622851, -0.021538189, -0.0047120233, 0.014083955, -0.021181997, -0.0066415817, -0.0078251185, 0.061450556, 0.021893805, -0.012969906, 0.028493797, 0.025576953, -0.017138785, 0.042083807, 0.00036911172, -0.030676272, 0.050215244, 0.015479565, -0.04796683, 0.033385694, -0.02503968, -0.05061087, -0.0491622, -0.02282723, 0.027421778, -0.022742985, 0.025241354, 0.010889715, 0.06654725, 0.0049236026, -0.031101132, 0.03536845, 0.0014072601, 0.03913315, -0.008463993, 0.009379072, 0.0006677232, -0.0009472659, -0.027066972, -0.005452731, -0.01926149, -0.007412688, -0.03445318, -0.0065549063, -0.025225064, -0.03491155, -0.0446118, -0.017254617, 0.060066648, 0.010801351, -0.009364108, -0.0144907, 0.026756773, -0.0057106786, -0.06387333, -0.05528889, -0.026777834, 0.029067881, 0.04143424, 0.029176494, 0.019399244, 0.014605724, 0.01621678, 0.03861098, 0.074201494, 0.0028264967, 0.02736897, -0.014865198, -0.043763593, 0.0273201, -0.023304801, 0.0017516458, -0.032924797, -0.010752354, 0.029936843, 0.021415757, 0.028821968, 0.019492527, -0.07589984, -0.024474056, 0.016437879, 0.035193823, -0.017879637, -0.0012485086, 0.022728937, -0.0066974284, -0.0593521, 0.010687849, -0.008661119, 0.002465336, -0.035397932, -0.024833398, -0.027838767, 0.04858304, -0.030275954, -0.059542, -0.014402102, 0.0039849435, 0.03153429, -0.013044847, -0.0026132346, -0.026405718, 0.024847649, 0.00024627347, -0.049437426, 0.05496891, 0.028762143, -0.024671769, 0.0451587, 0.0013513307, -0.007984648, 0.009694945, -0.027586076, -0.050622508, 0.014362672, 0.009058755, 0.0051575615, 0.0023077175, 0.039088514, 0.022634061, 0.0044931322, -0.03056379, -0.01936727, 0.011267317, 0.005696432, 0.009978829, -0.023386464, -0.07942381, -0.019201465, 0.032288373, -0.044093903, 0.005352696, -0.005885503, -0.0047914796, -0.024405414, 0.006466358, -0.043255966, 0.038595777, -0.006288799, 0.003101555, -0.00892414, 0.012949309, 0.035286594, 2.9721132e-05, 0.052450377, -0.0038501786, 0.0066879806, 0.008009971, 0.039799966, 0.06371628, 0.037566163, -0.0048061563, -0.029491583, -0.004748476, 0.01175604, -0.042217344, 0.039030153, -0.01766552, -0.013161762, -0.001503679, -0.010464876, 0.0067881905, 0.06120133, 0.049402766, 0.0019048704, 0.0084636565, 0.0114878295, -0.05378138, -0.03149985, -0.009728915, -0.0061442517, 0.023939742, 0.026228094, -0.06536439, -0.0070977937, -0.01675599, 0.020468276, 0.016525885, -0.02310716, -0.0047005336, 0.0024603687, 0.0015534266, 0.10586247, -0.030924536, 0.06698407, 0.024861407, -0.0201981, -0.019013833, 0.054178648, -0.032924637, 0.03400909, -0.013993611, 0.00044071834, 0.000559415, -0.021633198, 0.015791995, -0.006407045, 0.038585182, -0.044887774, 0.009475685, 0.030613726, -0.0113635985, 0.023756778, -0.014984713, 0.049845427, 0.010304104, -0.086357705, -0.016592525, -0.08997873, 0.018802254, -0.026621897, 0.036119964, -0.017771624, 0.00059697503, -0.02346701, -0.014322784, -0.008580323, 0.0069213705, 0.011175727, 0.0046809004, 0.038701963, -0.09481818, 0.011147942, 0.041168302, 0.014526665, -0.008420172, 0.011484339, 0.027245803, -0.016563807, 0.014359836, 0.0149701955, -0.0051656384, 0.006729749, -0.011603818, 0.0212209, -0.044899914, 0.016950674, 0.029617267, -0.0021991997, -0.008901735, 0.02886822, -0.11489113, -0.041502293, -0.008972503, -0.018324632, 0.020270532, 0.008680867, 0.014061169, -0.025777379, -0.008556963, 0.020973902, 0.012923231, -0.02128589, 0.004063174, -0.024218658, -0.0150889335, 0.011989689, -0.052606042, 0.033361696, -0.017586827, -0.035884257, -0.061752904, 0.0002652234, 0.01927769, -0.04834388, -0.0033916908, 0.016262652, -0.04296916, -0.0126989335, 0.060400326, -0.004198102, -0.02059228, 0.038904168, -0.0043159085, 0.018626427, 0.0019096491, 0.016388686, 0.005415011, -0.008333832, 0.017366035, -0.042109255, 0.038120534, 0.10888748, 0.008978477, -0.028394049, 0.07174635, 0.018888742, -0.01213845, 0.0025502867, 0.040750608, -0.016777312, -0.022716658, -0.039629992, -0.015548788, -0.035059884, -0.0076331603, -0.005629379, -0.004647178, -0.018005254, 0.0015239891, 0.045861274, 0.016498426, -0.032213043, 0.036940183, 0.081857264, -0.060237546, -0.0037613905, -0.0018755265, 0.049716476, -0.006466456, -0.004143031, -0.035811506, 0.020899538, 0.0101147825, -0.0036764666, -0.013907347, -0.0679712, 0.024115274, -0.037679743, 0.028760703, -0.0037167303, 0.043736957, 0.014521639, -0.003778043, -0.017076397, -0.0130539015, -0.038074847, -0.0043564457, 0.030681409, -0.005815582, -0.022514924, -0.010933786, 0.004588307, 0.0026248787, 0.004284429, -0.03298058, -0.015521615, 0.0064100004, -0.01893799, 0.0060417894, 0.018334668, -0.02198147, 0.027333694, 0.04018599, 0.021325776, -0.0025792646, -0.0070964796, -0.031838402, 0.044081926, 0.04020538, 0.013274614, -0.0067229043, -0.015172114, -0.020847015, -0.00855432, -0.028211147, -0.0069437996, -0.05849252, -0.044075612, 0.022457095, 0.028858433, -0.011914434, 0.004135809, 0.011121238, -0.005872629, -0.047949806, -0.06383951, -0.018321758, -0.0061903964, -0.011756049, -0.017300788, -0.02058743, -0.005997279, -0.041570965, -0.01887957, 0.031555332, 0.0027844524, -0.040020343, -0.0020960069, 0.023028765, -0.014180706, -0.017217258, 0.015452998, -0.029311024, 0.0006876796, -0.03474882, -0.30907038, -0.0058631236, 0.0014001256, -0.027516108, -0.007420682, -0.019168502, -0.012166478, 0.030878594, -0.01918663, 0.056898993, 0.05044068, 0.014607676, 0.014405426, -0.03685031, -0.009221327, 0.044843446, -0.01992298, -0.037504435, -0.035695698, 0.07108686, -0.0024651068, 0.015513091, -0.011931885, -0.058694184, 0.067271374, -0.028939627, 0.04480718, -0.025160747, 0.04374399, 0.014101961, 0.003215209, -0.034264565, 0.007925918, -0.006573396, -0.024641855, -0.020500943, 0.0071604154, -0.0064163352, -0.026567247, 0.0032201773, -0.0009257711, 0.030684479, 0.024450427, -0.019511536, -0.054222975, -0.061622933, -0.006927705, 0.030600494, -0.022812206, 0.018437557, -0.0036987814, -0.0027277123, -0.0060068737, 0.015108217, 0.07632971, -0.002774065, 0.062745616, -0.028119259, 0.02760193, 0.028346397, 0.034770597, 0.013260941, -0.021104988, -0.019841032, -0.0036810725, -0.008130473, -0.005543933, -0.007989633, 0.01971356, 0.0013070845, -0.04541393, -0.03849327, -0.0070400555, 0.039451294, 0.012223503, -0.052652247, 0.021981126, -0.00021076435, -0.009386243, 0.019419711, -0.0012487499, -0.01904301, 0.02556265, -0.008128203, -0.003732209, 0.010064544, -0.05195107, -0.031213244, 0.022518957, -0.0026264188, -0.010448385, -0.018521685, 0.03493823, -0.008060339, 0.014538763, -0.028849382, 0.033285458, 0.028162101, -6.0668757e-05, 0.003654882, 0.01206097, -0.011459334, -0.03767873, -0.0050211097, -0.018558295, 0.025982862, -0.007263257, 0.08609022, 0.013457168, 0.033720117, -0.028466608, -0.056001462, 0.06382726, -0.00946123, 0.01337294, -0.009722221, 0.022457223, 0.035206154, -0.039755434, 0.017857952, -0.00020658875, 0.031036485, -0.0041659875, 0.044071283, -0.021184653, 0.036927965, -0.027687192, -0.036322925, -0.03465842, 0.0024703194, 0.023718681, -0.005140203, 0.10584172, -0.026125826, 0.019445334, 0.04961985, -0.0164236, 0.012177855, 0.001068047, 0.020617414, -0.018489014, -0.017424006, -0.03253692, -0.028224336, -0.037873995, -0.00041655678, -0.000543745, 0.01819874, -0.019409256, -0.03748955, 0.00056815933, -0.019601634, -0.058148917, 0.006333733, 0.013043057, 0.022930855, 0.026059251, -0.06731322, -0.12764573, -0.013037023, 0.03711542, -0.044639062, -0.033755966, 0.008360664, 0.04712386, 0.0043586367, 0.0021042095, -0.013222326, 0.0140003385, -0.0023824326, 0.03345798, -0.015407794, 0.017776951, 0.016651254, -0.008489233, -0.018615602, -0.020112548, -0.01109594, -0.012095163, -0.011609609, 0.021557726, -0.02015948, -0.036674663, 0.0028227586, 0.011397878, 0.010906482, 0.0056243674, 0.0067206807, 0.03346026, -0.007995403, -0.018333195, 0.061646536, -0.03148897, -0.0059196264, -0.046350162, 0.03869765, -0.0141338585, 0.01783148, -0.005160642, 0.02233562, 0.02824081, 0.040886793, 0.016690927, -0.019632982, 0.0034978963, 0.025622422, 0.029175192, 0.03129038, 0.007965436, -0.014479083, -0.029063404, 0.002176511, -0.025867373, -0.014531188, 0.0194007, -0.016332185, 0.0085755065, -0.01607479, 0.014753511, -0.00031572027, 0.05080484, -0.013936455, 0.046942055, 0.004900548, 0.010416213, 0.013098065, 0.006103205, 0.0050759623, 0.028589582, 0.009013781, 0.011409561, -0.013666879, -0.07381896, -0.020935267, -0.016496032, -0.090582944, 0.019194711, 0.0029344114, 0.014002559, -0.0064830654, 0.038881477, -0.009100745, -0.03219273, 0.02314725, -0.014437725, -0.04093972, 0.03271461, -0.037313137, 0.023839032, -0.007007703, 0.0378476, 0.015339351, 0.046282712, -0.020478265, -0.045257237, -0.010490222, -0.011947393, -0.042391088, -0.018853765, 0.0033090068, 0.024678256, 0.04145972, 0.025309248, 0.028540693, 0.008458297, -0.04335141, 0.024804942, -0.0028688177, -0.03370457, 0.0083794715, -0.028840652, 0.013122521, -0.0037777252, -0.091269724, -0.020235308, 0.008682096, 0.008378103, -0.0037765398, 0.034532033, 0.011649039, -0.023977311, 0.028237429, -0.004338265, 0.02215903, 0.008142726, 0.04325965, 0.025667787, 0.04021542, 0.022507671, -0.0064422973, 0.028300997, -0.012631801, 0.0075581823, -0.0172389, -0.016823545, -0.025513537, 0.032608386, 0.036558542, 0.021230185, 0.018272884, -0.003206782, -0.001513966, -0.009737264, 0.052647226, 0.004755796, -0.012751857, 0.0061726454, -0.019755648, -0.031904165, 0.032670375, -0.03168409, 0.0014640073, -0.048017967, 0.033818897, -0.010460395, 0.04110712, 0.018695794, -0.03581825, -0.019928379, 0.018624658, 0.0136633245, -0.0018563502, 0.020398887, 0.0039079227, -0.018455623, -0.042622063, -4.808325e-05, -0.0059154034, 0.010688947, -0.009454969, -0.01601241, -0.017247252, -0.04350985, -0.052176755, -0.026783822, 0.011961513, 0.10685226, 0.0094851, 0.004242087, 0.0014396007, 0.02782559, 0.00833203, -0.040681683, -0.0036343292, -0.0013086532, -0.004013325, -0.01698305, -0.003161006, 0.014381047, -0.0055312733, 0.017707074, -0.008367256, -0.034846142, 0.027633574, 0.042642724, 0.0038731259, 0.021297354, -0.020382486, -0.034834724, 0.0079073245, 0.03058069, -0.02682858, -0.12922525, 0.043005534, -0.028537868, -0.032668922, 0.010318783, 0.009427436, -0.014817888, -0.022491561, 0.045085978, 0.016351378, 0.060069792, 0.02156117, 0.007826801, 0.032951668, -0.013345699, -0.0491277, -0.067106046, -0.025947224, -0.0112283565, -0.002680028, 0.030128973, 0.0075812, -0.0041002613, -0.05550901, 0.006174435, 0.060282163, -0.015323213, 0.023874965, 0.020932458, -0.040702835, -0.0016292671, -0.008365611, -0.014588778, -0.01889341, 0.013125672, -0.021439059, 0.026925864, -0.012446013, 0.011465863, 0.07228482, -0.03347981, -0.01949296, 0.006473508, -0.03492369, -0.005713642, 0.027152188, 0.020540444, -0.009710128, -0.030627985, 0.017034076, 0.016081078], &#39;start_time&#39;: 12.0, &#39;end_time&#39;: 18.0, &#39;embedding_scope&#39;: &#39;clip&#39;}<br>  Scope: clip<br>  Time range: 12.0 - 18.0 seconds<br>  Embedding vector (first 5 values): [0.022766022, -0.0035799714, 0.019664006, -0.023654077, -0.003662397]</pre><p>You can clearly see a long video has been broken into short clips as well as the embedding vector.</p><p><strong>Step 5: Insert vector into Milvus</strong></p><pre># Insert text and all video segment embeddings<br>def insert_embeddings(milvus_client, collection_name, video_embeddings, video_segments):<br>    video_entry = []<br><br>    # Insert each video segment embedding<br>    id = 0<br>    for segment in video_segments[0:]:<br>        print(type(segment.embeddings_float))<br>        print(type(segment.start_offset_sec))<br><br>        segment.embedding_scope<br>        video_entry.append({<br>            &#39;id&#39;: id,<br>            &#39;vector&#39;: segment.embeddings_float,<br>            &#39;start_time&#39;: segment.start_offset_sec,<br>            &#39;end_time&#39;: segment.end_offset_sec,<br>            &#39;embedding_scope&#39;: segment.embedding_scope<br>            <br>        })<br>        id += 1<br>        <br>        insert_result = milvus_client.insert(collection_name=collection_name, data = video_entry)<br><br>insert_result = insert_embeddings(milvus_client, collection_name, video_embeddings, video_segments)</pre><p><strong>Step 6: Search for similar videos using text query</strong></p><p>Now it is the fun part. Imagine if a SMB owner wants to run a marketing campaign during the holiday season, they can simply search video caption, tags, metadata that might includes a certain keyword (e.g: “holiday season”, “summer vibe”, etc), they can simply do that with the embedding we just extracted from the videos.</p><pre># Search for similar video segments using text query<br><br>def perform_similarity_search(milvus_client, collection_name, query_vector, limit=5):<br>    &quot;&quot;&quot;<br>    Perform a similarity search on the Milvus collection.<br><br>    Args:<br>        milvus_client: The Milvus client instance.<br>        collection_name (str): The name of the Milvus collection to search in.<br>        query_vector (list): The query vector to search for similar embeddings.<br>        limit (int, optional): The maximum number of results to return. Defaults to 5.<br><br>    Returns:<br>        list: A list of search results, where each result is a dictionary containing<br>              the matched entity&#39;s metadata and similarity score.<br><br>    This function searches the specified Milvus collection for embeddings similar to<br>    the given query vector. It returns the top matching results, including metadata<br>    such as the embedding scope, time range, and associated video URL for each match.<br>    &quot;&quot;&quot;<br>    search_results = milvus_client.search(<br>        collection_name=collection_name,<br>        data=[query_vector],<br>        limit=limit,<br>        output_fields=[&quot;embedding_scope&quot;, &quot;start_offset_sec&quot;, &quot;end_offset_sec&quot;, &quot;video_url&quot;]<br>    )<br><br>    return search_results</pre><p><strong>Step 7: get embedding from your search text</strong></p><pre># Function to get embedding for a text query<br>def get_text_embedding(text_query):<br>    # Twelve Labs Embed API supports text-to-embedding<br>    text_embedding = twelvelabs_client.embed.create(<br>      model_name=&quot;Marengo-retrieval-2.7&quot;,<br>      text=text_query,<br>      text_truncate=&quot;start&quot;<br>    )<br><br>    return text_embedding.text_embedding.segments[0].embeddings_float</pre><p>Note: you need to use the same model as your video. In this case, “Marengo-retrieval-2.7&quot;.</p><p><strong>Step 8: Let’s test it out!</strong></p><pre># define the query vector: text-based search<br>text_query = &quot;A big white rabbit is walking in the forrest. &quot;<br>query_vector = get_text_embedding(text_query)<br><br># Perform a similarity search on the Milvus collection<br>search_results = perform_similarity_search(milvus_client, collection_name, query_vector)<br>print(search_results)</pre><pre>data: [&quot;[{&#39;id&#39;: 10, &#39;distance&#39;: 0.39097675681114197, &#39;entity&#39;: {&#39;embedding_scope&#39;: &#39;clip&#39;}}]&quot;]</pre><p>I tested a white rabbit only because I got a cute rabbit in my video example. 🐰</p><p>Oh hey, there is also a cat in the video. 😹</p><pre># define the query vector: text-based search<br>text_query = &quot;I am a grumppy cat.&quot;<br>query_vector = get_text_embedding(text_query)<br><br># Perform a similarity search on the Milvus collection<br>search_results = perform_similarity_search(milvus_client, collection_name, query_vector)<br>print(search_results)</pre><pre>data: [&quot;[{&#39;id&#39;: 36, &#39;distance&#39;: 0.23986005783081055, &#39;entity&#39;: {&#39;embedding_scope&#39;: &#39;clip&#39;}}]&quot;]</pre><p>Well, it is from a different clip. The similarity score is a bit low and I’m not sure if it is because the cat is grumpy or not.</p><p>That’s it! Even though this is not a production-ready RAG, but I definitely learned a ton. I will definitely explore some deploy tools to play with and see how I might continue to improve it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6287b79ddc10" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Spinnaker green/blue deployment cleanup solution]]></title>
            <link>https://lijie-zhou.medium.com/spinnaker-green-blue-deployment-cleanup-solution-9673ee110374?source=rss-e44b1724aa04------2</link>
            <guid isPermaLink="false">https://medium.com/p/9673ee110374</guid>
            <category><![CDATA[spinnaker]]></category>
            <category><![CDATA[infrastructure]]></category>
            <category><![CDATA[deployment]]></category>
            <category><![CDATA[kubernetes]]></category>
            <category><![CDATA[devops]]></category>
            <dc:creator><![CDATA[Lijie Zhou]]></dc:creator>
            <pubDate>Sat, 02 Jan 2021 02:59:50 GMT</pubDate>
            <atom:updated>2021-01-02T02:59:50.504Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/633/1*WfB4HxWJH_02iO7hYeCSIA.png" /></figure><p>Earlier this year, we adopted <a href="https://spinnaker.io/">Spinnaker</a> as our deployment tool at Gusto. By enabling the <a href="https://martinfowler.com/bliki/BlueGreenDeployment.html">green/blue deployment</a> on top of the core Kubernetes infrastructure, we have seen a big improvement in terms of the deployment time. Thus, this is definitely a big win for us.</p><p>However, Spinnaker <a href="https://spinnaker.io/guides/user/kubernetes-v2/traffic-management/#you-must-use-replica-sets">uses ReplicaSet to manage its blue/green deployment</a>, which brought us a new set of challenges in managing the deployment pipeline. In Kubernetes, <a href="https://kubernetes.io/docs/concepts/workloads/controllers/replicaset/#:~:text=A%20ReplicaSet%20ensures%20that%20a,lot%20of%20other%20useful%20features.">ReplicaSet is a lower abstraction relative to deployment</a>, which means Kubernetes will not be able to clean up the ReplicaSet directly if something goes wrong at the ReplicaSet level. What could go wrong? Let’s start with a simple deploy pipeline.</p><h3>A simple deploy pipeline</h3><p>Let’s say we have a simple deployment pipeline that includes three services: WebApp, Stream Consumer, and Workers. All services bake their manifests and deploy simultaneously. We enable the traffic on the new version (candidate), disable traffic on the old version (incumbent), and delete the old version of ReplicaSets.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*c6fZh-OFtylheEUuw7abNw.png" /><figcaption>A simple deploy pipeline deploying three services</figcaption></figure><p>Before the step of “Enable (Candidate)”, the state of the pipeline is the following</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/675/1*PyTwirDZQYD0Zex2t9QYEA.png" /></figure><p>If everything works as expected, the following will happen:</p><ul><li>The “Enable (Candidate)” step will add a label to the new versions of ReplicaSets to enable the traffic of all the services.</li><li>The “Disable (Incumbent)” step will remove the label that was added from the previous successful deployment.</li><li>After the traffic is disabled, the old ReplicaSets will be deleted. Assuming the “disable (incumbent)” step succeeded, no service disruption will incur.</li></ul><h3>Multiple ways to go wrong</h3><p>In reality, we have observed several scenarios that can cause the pipeline to fail at different stages, and here are the states of deploy pipeline when they fail.</p><ul><li>Scenario 1: Deploy fails at the deploy stage</li><li>Scenario 2: Deploy fails at the enable candidate stage</li><li>Scenario 3: Deploy fails at the disable/delete incumbent stage</li></ul><h3>Scenario 1</h3><p>Occasionally, there is a bad deployment that “bypasses” the <a href="https://kubernetes.io/docs/tasks/configure-pod-container/configure-liveness-readiness-startup-probes/">basic Kubernetes health check</a>. In this case, the bad deploy will get stuck on the “deploy” stage until it passes the deploy deadline. The traffic is still on the old versions because the new version’s traffic has not been enabled yet.</p><p><em>(I use “T” to mark the ReplicaSets that currently have traffic on)</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/677/1*PswQguN2e_nYwj2IK9D0Ig.png" /></figure><h3>Scenario 2</h3><p>This is a very rare case but we did see this happen once or twice. For whatever reason, when Spinnaker tries to put a label on the new version ReplicaSets, it failed. This is a very messy and dangerous case because you have both old and new versions serving traffic in a failed state.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/676/1*OhzIfy_ns_aBYhdSzyCdDA.png" /></figure><h3>Scenario 3</h3><p>There is a slim chance that Spinnaker tries to disable traffic on a host that Kubernetes tries to scale down. If that happens, Spinnaker does not “know” the host is actually gone and still tries to remove the label which will never succeed. In this case, the newer version is already serving traffic. Thus, we can configure the Spinnaker pipeline to ignore the failure and continue to delete the old ReplicaSets.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/681/1*ORVmdgpz1-2hCmKRqol3Vg.png" /></figure><p>Because Kubernetes will not clean up these ReplicaSets, we will need to figure out a solution to handle each of the failure states. Additionally, timing is important. It is better to do this right after the pipeline failed to avoid multiple failed versions to stack up (or even worse, the old version still serves traffic). The solution we have chosen is: <strong>find the most updated version with the traffic and keep that. Delete the rest of the ReplicaSets. This will get the pipeline back to a healthy state in each of the scenarios listed above.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*----DKFESNe9-cA9qa_uww.png" /><figcaption>A pipeline with the cleanup step</figcaption></figure><p>We can easily write a python script to achieve this goal. This solves the first challenge of this problem. But the next question is how to invoke this script and pull the ReplicaSet objects from the Spinnaker pipeline?</p><h3>How to run the cleanup script safely?</h3><p>The first thing I have considered is <a href="https://spinnaker.io/setup/features/script-stage/">“Run script” via Spinnaker.</a> The Spinnaker <a href="https://spinnaker.io/reference/pipeline/stages/#script">“Script stage” lets you</a> run an arbitrary shell, Python, or Groovy script on a Jenkins instance as the first-class state in Spinnaker. However, it will be extra maintenance to configure Jenkins hosts to do that.</p><p>The next thought is to use Spinnaker’s <a href="https://spinnaker.io/reference/pipeline/stages/#run-job">run job</a> to run a container. If that works, I just need to <a href="https://www.pybootcamp.com/blog/how-to-write-dockerfile-python-apps/">Dockerize the Python script</a> and <a href="https://aws.amazon.com/ecr/">push it to ECR</a> for Spinnaker to pull. Here is the example Manifest of the Spinnaker job:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zpM8Ta3DZJMcV4neGRZ4pQ.png" /><figcaption>A sample manifest of the Spinnaker job</figcaption></figure><p>Additionally, we will need to <a href="https://itnext.io/running-kubectl-commands-from-within-a-pod-b303e8176088">create a service account </a>for the “internal pod” to run the job (t<a href="https://kubernetes.io/docs/tasks/access-application-cluster/access-cluster/#accessing-the-api-from-a-pod">he “internal pod” approach is a very common way to configure custom logic within the cluster).</a> We will need to give the “get” and “list” permission to both “Pods” and “ReplicaSets”.</p><pre>apiVersion: rbac.authorization.k8s.io/v1<br>kind: Role<br>metadata:<br>  name: modify-pods<br>rules:<br>  - apiGroups: [&quot;apps&quot;]<br>    resources:<br>      - pods<br>      - replicasets<br>    verbs:<br>      - get<br>      - list</pre><p>One small but important detail is how to make the Python script to talk to the Kubernetes cluster. The simplest way is probably to <a href="https://janakiev.com/blog/python-shell-commands/">execute a Shell command and there are multiple ways</a> to do just that. Alternatively, you can also use the <a href="https://github.com/kubernetes-client/python/blob/master/kubernetes/README.md">Kubernetes Python client</a> which is probably more robust. This library is not very well documented so you may run into some confusion when calling certain APIs. Here is a code snippet to explore if you follow the “agent” pod approach.</p><pre>config.load_incluster_config()<br>    v1 = client.AppsV1Api()<br>    ret = v1.list_namespaced_replica_set(&quot;example&quot;)</pre><h3>Further thoughts</h3><p>In Kubernetes’ world, this starts to feel like a “too many chefs in the kitchen” scenario. There is a certain risk associated with running this job because it may accidentally delete something. If you are using Spinnaker to run blue/green deployment and have similar issues, I’d be interested to learn what you do to solve this problem.</p><p>(Thanks <a href="https://medium.com/@mallyvai">Vaibhav Mallya </a>for proofreading the post)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9673ee110374" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Two Years As A SRE]]></title>
            <link>https://lijie-zhou.medium.com/two-years-as-a-sre-8ab24abedfb7?source=rss-e44b1724aa04------2</link>
            <guid isPermaLink="false">https://medium.com/p/8ab24abedfb7</guid>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[site-reliability-engineer]]></category>
            <category><![CDATA[startup-life]]></category>
            <category><![CDATA[career-development]]></category>
            <dc:creator><![CDATA[Lijie Zhou]]></dc:creator>
            <pubDate>Mon, 02 Nov 2020 19:15:20 GMT</pubDate>
            <atom:updated>2020-11-03T06:01:47.739Z</atom:updated>
            <content:encoded><![CDATA[<p>Last month, I passed my two year anniversary working as a Site Reliability Engineer (SRE) at <a href="https://gusto.com/">Gusto</a>. Several people asked me on Quora <a href="https://www.quora.com/What-is-SRE-What-do-they-do">what do SREs do</a> and <a href="https://www.quora.com/Why-would-somebody-become-a-Site-Reliability-engineer-at-Google-instead-of-a-software-engineer-They-have-burdensome-on-call-shifts-and-dont-directly-work-on-the-core-product">why do people want to become a SRE instead of SWE focusing on building features</a>. I think this is a good time to reflect and share the knowledge that I have learned in the past two years.</p><p><strong>My background</strong></p><p>I previously worked as a security engineer at Facebook and McAfee before I started my SRE role at Gusto. Actually, I was hired as a security engineer when I joined. My manager at that time recommended me to spend my first six months on the infrastructure team to get a big picture of how infrastructure works and I agreed. That was the best decision I’ve ever made (thank you <a href="https://medium.com/u/5b1a9d44c16f">Jack Danger</a>) because I quickly fell in love with what I do. I used to struggle with which area to focus on in security because it is such a big area: application security, detection &amp; response (D&amp;R), security operation, malware and reverse engineering, etc. No one can be an expert on all of them, but I found it hard to commit myself into either of these deeply.</p><p><strong>Infrastructure is about system thinking</strong></p><p>As an infrastructure engineer (a.k.a devops), you need a good sense of each component in the deployment system and how they work together. At Gusto, we are the first tier responders to both internal and external services. Different from big companies which may have different groups to be responsible for different areas, we are responsible for most of the services after the developers push their code to production. This means you need to at least have working knowledge of the following stack of technologies:</p><ul><li>Docker</li><li>Linux</li><li>Deploy Pipeline: Jenkins, Buildkite, Spinnaker</li><li>Kubernetes</li><li>AWS services (EC2, Autoscaling, DNS, RDS, VPC, S3, SSM, etc)</li><li>Infrastructure-as-code tooling (Terraform)</li></ul><p>Usually, the wide range of technologies for an infrastructure engineer to learn is overwhelming. Different from SWE, who usually transition from one language to another, it usually takes years to become a good infrastructure engineer because it takes a long time to learn all the system layer technologies.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*WB7tVC0r9d9L6WrAUVKPLQ.jpeg" /></figure><p><strong>Deep dive into a specific domain when necessary</strong></p><p>Whenever necessary, it is very important to quickly dive deep into a specific domain. For people who use <a href="https://www.terraform.io/">Terraform</a> to manage their infrastructure-as-code, you must know that there are many quirkiness of Terraform. For example, if you want to declare a <a href="https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/security-group-rules-reference.html">AWS security group rule</a>, you can either declare it as a stand-alone resource or in-line rule within the aws_security_group itself. <a href="https://registry.terraform.io/providers/hashicorp/aws/latest/docs/resources/security_group">However, you cannot mix them both.</a> The consequence of having the mixed ways of managing security group rules is that Terraform will destroy the rules and recreate them. Thus, even though there may be no actual change on the security group rules in AWS, your service may experience disruption during the Terraform destroy/recreate process. I learned this lesson in a hard way.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*BaLM5c7MydPCaMS27BqvFA.jpeg" /></figure><p><strong>Learn by doing</strong></p><p>There is no infrastructure-101. Usually, infrastructure is not an entry-level role because of all the reasons mentioned above. No computer science program teaches infrastructure-101 either. Thus, the best way to learn is by doing. Register an AWS free account and launch an EC2 to start with (that’s how I got started). You may not have all the scaling challenges confronted by most startups, but that is a good starting point to learn the basics.</p><p>Also, the speed of new tools and technologies in this domain is mind-blowing. Since I joined Gusto, we migrated our core infrastructures on <a href="https://kubernetes.io/">Kubernetes</a> and adopted <a href="https://harness.io/build-pipelines-in-minutes/?utm_source=google&amp;utm_medium=cpc&amp;utm_campaign=search-spinnaker&amp;utm_content=us-en-trial-start-your-free-trial&amp;utm_term=spinnaker-eta&amp;gclid=CjwKCAiA-f78BRBbEiwATKRRBCz-lxI6UDCFZuoiI8vGVXEe-QNE2VwD2bamPo5MiAYjQTX9Ot-8wRoCKtwQAvD_BwE">Spinnaker</a> as our deployment pipeline to better suit Kubernetes. Looking back, it is hard to believe I had zero knowledge about Kubernetes 18 months ago. If you love learning new things, this field will not get you bored because you will always have new things to learn.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/638/1*AYD12s_L8FrGoNJ1r9-iRw.png" /></figure><p><strong>On-call and incident response</strong></p><p>A lot of people ask me what on-call feels like and how often do I need to lead an incident. Honestly, on-call is draining, especially if you get an unlucky week. However, it is kind of like a muscle that you can train yourself, both professionally and mentally.</p><p>First of all, you will build the intuition what could go wrong after you get familiar with the system. For example, if your application is experiencing slowness, a list of things to check is: did your autoscaling scale fast enough relative to your incoming web requests? did your autoscaling group launch successfully? how is your DB performance? In reality, each of these core piece should have some monitoring/alerting in place and the alarm should fire up when passing its normal benchmark. Most companies also have an on-call runbook to give you a pointer for common incident issues and its resolution.</p><p>Second, think of the on-call person as a router plus a problem identifier. At Gusto, we do not expect the on-call person to give answer to all the questions simply because there are so many changes going on. Thus, you need to know your team well and find the right person after you identify the right problem. That encourages engineers to do pair programming and share knowledge, which consequently build strong trust within the team.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*9QCWck6u7aSV_ir8ou6oLQ.jpeg" /></figure><p>There is probably a lot more I can talk about and I will try to write more in the future. If there is any specific question you want me to answer, feel free to ask in the comment.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8ab24abedfb7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Women in Cybersecurity 2018 Recap]]></title>
            <link>https://lijie-zhou.medium.com/women-in-cybersecurity-2018-recap-349960d2d814?source=rss-e44b1724aa04------2</link>
            <guid isPermaLink="false">https://medium.com/p/349960d2d814</guid>
            <category><![CDATA[women-in-tech]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <dc:creator><![CDATA[Lijie Zhou]]></dc:creator>
            <pubDate>Mon, 16 Apr 2018 23:24:23 GMT</pubDate>
            <atom:updated>2018-04-16T23:24:23.913Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hE44EfT45-oNlDaMN6SqLQ.jpeg" /><figcaption>2018 Women in Cybersecurity (WiCyS) Chicago</figcaption></figure><p>I had an amazing experience last month in Chicago to attend 2018 <a href="https://www.wicys.net/">Women in Cybersecurity conference</a>. I got to meet more women and learn about their work. I respect every single one of them by their courage, technical leadership, and commitment to getting more women into this man dominated field. There are so many things that I learnt in the fleeting two days, but I will try my best to summarize them into three main things:</p><h3><strong>Stand up and present your work</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IkCCIV9WFSzMYqm2s_xdxg.jpeg" /><figcaption>Me presenting my work in Twitter spam detection at WiCyS</figcaption></figure><p>I went to WiCyS 2017 as a participant. At that time, I was too scared to even submit a proposal. However, when I left Arizona last year, I decided that I’m going to present my work next year in Chicago. There are many times I am ignored at security conferences as a woman, and I choose not to opt myself out this time. I was so lucky to be selected as one of the few <a href="https://www.slideshare.net/Annelijie/a-deep-learning-approach-for-twitter-spam-detection-lijie-zhou">technical presentation</a> speakers this year and I loved every minute of it when I shared my work with the audience. Before this presentation, I thought I am much more comfortable to share my ideas with others on Github or Medium, but I was wrong. I learnt I just need to push myself harder to get out of my comfort zone.</p><h3>Get a male mentor/ally</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UubxqmTUnH8kXqamjffxEw.jpeg" /><figcaption>Panel discussion with Ben Hogen (Facebook), Chintan Parekh (Fidelity), <em>Rodney Petersen (NIST), Greg Shannon(</em>Carnegie Mellon University<em>), </em>Tony Baylis (Lawrence Livermore National Laboratory) and <em>Tony Jeffs (Cisco)</em></figcaption></figure><p>This year, there is a new panel discussion about how to find a male mentor/ally and learn things from a different perspective and I absolutely love it. I have observed this pattern that I usually get different career advice from male and female mentors and that difference really helped me to find a unique way to solve the problem. I learned that when you are in a less female-friendly environment, instead of trying hard to fit in, you should always find someone that you trust and get his support to move forward. I want to give a big shoutout to the men who care about diversity in security and are committed to make a difference.</p><h3>There are so many new challenges and we need to figure out together</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KL0IY_2h0_KqypAz3XD6Aw.jpeg" /><figcaption>Amazing poster sessions</figcaption></figure><p>I learned so much by talking to people at the conference. It is impossible for me to learn everything in security and this field is evolving every day. I realize how important it is to grasp knowledge by talking to others. From this perspective, networking and communication are really important skills to make you successful. <strong>Security is not some hackers sit in the darkness trying to hack into someone’s computer. In my opinion, it is people from different backgrounds who can contribute their unique background and skillset to jointly solve a problem.</strong></p><p>Last but least, I am building a Youtube Channel to interview Chinese Women in Security. I will post our first video soon. At the same time, if you are interested or have any candidate to recommend, I would really appreciate.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=349960d2d814" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[WordPress Blog Got Hacked]]></title>
            <link>https://lijie-zhou.medium.com/wordpress-blog-got-hacked-ccb3bfdad9c2?source=rss-e44b1724aa04------2</link>
            <guid isPermaLink="false">https://medium.com/p/ccb3bfdad9c2</guid>
            <category><![CDATA[wordpress-security]]></category>
            <category><![CDATA[security]]></category>
            <category><![CDATA[wordpress-plugins]]></category>
            <category><![CDATA[wordpress]]></category>
            <dc:creator><![CDATA[Lijie Zhou]]></dc:creator>
            <pubDate>Fri, 16 Mar 2018 02:33:28 GMT</pubDate>
            <atom:updated>2018-03-16T04:35:49.810Z</atom:updated>
            <content:encoded><![CDATA[<p>I have been running a WordPress (WP) site as a side project for a while. This week, when I opened my site, I found that the site has several blogs that I have never posted. I realized <strong>I may have been hacked</strong>!</p><p><strong>Assumption Confirmed</strong></p><p>In order to prove my guess, I logged into my recent access log and found a large amount of bot records.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*v-1Ub8euONiGKAAZC1g-hA.png" /><figcaption>This is a small portion of the bot records</figcaption></figure><p><a href="https://aw-snap.info/">Aw-Snap</a> also confirms that my site is probably infected by the <a href="https://aw-snap.info/articles/spam-hacks.php">spam hack</a>.They even give me the spammy words that hackers embedded into my site (essay|writing|cheap).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rvnelVnMIoesF4TpfexgaQ.png" /><figcaption>Source: Aw-Snap</figcaption></figure><p>At this point, I am 99% certain that my guess is correct. I use password manager for everything, so it is highly unlikely that the attacker got into my account due to weak password. Then, the vulnerability is more likely to come from plug-ins and/or theme I installed via WP.</p><p>I then contacted my host to see if they recently have compromised accounts at a large scale, and the answer is “no”. Thus, I documented every plug-in I installed and decided to recover the account next.</p><p><strong>Site Recovery</strong></p><p>Step 0: Find your backup (hopefully you have one!)</p><p>You should always backup your site after any update. There is <a href="http://www.wpexplorer.com/backup-wordpress-site-plugins/">a great article</a> about how to backup your WP site and it is really simple!</p><p>Step 1: Manually clean your site</p><p>I assume not every person have the backup option. In that case, you have to do the cleaning job manually. <a href="https://sucuri.net/">Sucuri</a> is a great tool that you can install directly from WP plug-in.</p><p>After installing Sucuri, use the most strict scanning level to do a comprehensive scanning. Sucuri will actually tell you what files have been modified.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1003/1*_5oUWjCZsDmelaepa-CEKA.png" /><figcaption>Sucuri Scanning Report</figcaption></figure><p>Go through each file and make your best judgement to decide whether it is a false positive or real malicious script. There are a lot of sample malicious scripts for your reference.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/754/1*9vLTiTC2mdJweXyXRdNSOQ.png" /><figcaption>These are the modified files on my webhost</figcaption></figure><p>Step 1.5: Do not trust one single tool. Try another one and scan again.</p><p>Wordfence is another WP security tool that has been around for a while. Choose the “High Sensitivity” option and do another scanning. Here is a nice <a href="https://stratusly.com/wordpress-security-sucuri-vs-wordfence-review/">analysis</a> of pros and cons of Wordfence and Sucuri.</p><p>Step 2: Change your password</p><p>Step 3: Upgrade your WP core, theme, and all the installed plugins</p><p>After this is done, you may want to do another backup.</p><p>Step 4: Setup defense mechanism</p><p>How do I know the attacker won’t attack my site again? Here are a few tips:</p><p>Tip 1: Setup daily activity email via either Wordfence or sucuri. Wordfence tells me the top10 failed logins, which means if the attacker tries to brute force into my account again, I will be notified immediately. Additionally, they also send notification for recently modified files.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/683/1*7RXCaJhyIaYHgii_9fc38A.png" /><figcaption>The attacker tried to login after I reset my password</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/676/1*cgQPfhEB-OBa_GoxZ_xQ7w.png" /></figure><p>Sucuri also sent me emails when the attacker tried to log into my account after I reset the password. Setup a filter in your mailbox if you don’t want to overwhelm yourself. ;)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/968/1*alZzRoPp6vMVOBlNvRHZ9Q.png" /></figure><p>Tip 2: Setup a Google alert to the same spammy word set on your site. If the attacker break into your site and post them again, at least you will be notified by Google instead of your customers.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ccb3bfdad9c2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Virtual credit card number and Google voice]]></title>
            <link>https://lijie-zhou.medium.com/virtual-credit-card-number-and-google-voice-90360c860932?source=rss-e44b1724aa04------2</link>
            <guid isPermaLink="false">https://medium.com/p/90360c860932</guid>
            <category><![CDATA[privacy]]></category>
            <category><![CDATA[information-security]]></category>
            <category><![CDATA[privacy-protection]]></category>
            <dc:creator><![CDATA[Lijie Zhou]]></dc:creator>
            <pubDate>Fri, 23 Feb 2018 18:00:19 GMT</pubDate>
            <atom:updated>2018-02-23T18:00:19.158Z</atom:updated>
            <content:encoded><![CDATA[<p>This is the second post of the personal digital <a href="https://medium.com/@lijie_zhou/google-knows-you-better-than-yourself-and-what-could-you-do-about-it-cb75c393d155">privacy</a> series. This post, I am going to share some tools I use to manage my personal digital privacy.</p><h3>Virtual Credit Card Number</h3><p>We are more and more dependent on online payment. However, I have never felt comfortable when giving my credit card information to some random website. There are some amazing <a href="https://stripe.com/">tools</a> out there for developers to integrate this functionality easier, which lowers the risk of credential leak due to bad implementation. However, small business still need to store users’ credit card info properly and that is no way for users to validate that.</p><p>I have two rules dealing with online payment:</p><p><strong>Rule 1: I have a (very) limited whitelist of companies that I use my credit card info.</strong></p><p>There is no guarantee that big companies won’t make mistakes (actually they messed up several times in the past, like <a href="https://www.theregister.co.uk/2017/12/04/paypal_tio_data_breach/">this</a> and <a href="https://www.wired.com/story/uber-paid-off-hackers-to-hide-a-57-million-user-data-breach/">this</a>). However, chances are much lower than small startups which do not even have a security team. Also, I have to use Amazon and I am not going to generate a virtual number every time when I use it.</p><p><strong>Rule 2: I use a virtual card number for transactions that are not from my whitelist.</strong></p><p>Most banks nowadays have the virtual card number as a FREE service, and it is very easy to use. All you need to do is to log in to your account and generate this one-time number and give it to the merchant. The downside is that if it is a recursive subscription, you have to do this every time you renew the service, and that is the decision you have to make for yourself between security and convenience. However, for any one-time payment, I see no reason to give your real credit card information.</p><h3>Google Voice Number</h3><p>Thanks Google for this awesome FREE service so that I do not need to give random people my real phone number. Google voice is basically another virtual number that you can use to add an extra layer of security to your personal phone number.</p><p><strong>Step 1: Go to </strong><a href="https://voice.google.com/"><strong>Google Voice</strong></a><strong> to register an account. Google will generate a number that looks exactly like a phone number for you.</strong></p><p><strong>Step 2: Link your real phone number with this Google voice number so that all the photo calls to your Google voice will be transferred to your real phone.</strong></p><p><strong>Step 3: Feel free to give people your google voice number.</strong></p><p>I really like Google voice because it has all the features that are like the real phone, such as: do not disturb, record a voice message, etc. I can also use “analyze the voice message” so that I only need to read the voice message transcript. There are two benefits of doing this: first, I can connect my Google voice to my email so that I only need to pay attention on my inbox. Second, automatically I backup all my call records and voice messages in my inbox.</p><p>I hope these two tips can inspire you to find more creative ways to protect your personal privacy. The simple rule is to add extra security layers on top of your most valuable information so that you can share wisely. I am sure there are other “privacy hacks” out there and I look forward to learning more.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=90360c860932" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Google knows you better than yourself and what could you do about it?]]></title>
            <link>https://lijie-zhou.medium.com/google-knows-you-better-than-yourself-and-what-could-you-do-about-it-cb75c393d155?source=rss-e44b1724aa04------2</link>
            <guid isPermaLink="false">https://medium.com/p/cb75c393d155</guid>
            <category><![CDATA[digital-identity]]></category>
            <category><![CDATA[privacy]]></category>
            <category><![CDATA[information-security]]></category>
            <dc:creator><![CDATA[Lijie Zhou]]></dc:creator>
            <pubDate>Fri, 29 Sep 2017 06:50:38 GMT</pubDate>
            <atom:updated>2017-09-29T16:35:04.159Z</atom:updated>
            <content:encoded><![CDATA[<p>Many people know that big tech companies know a lot about you, but few understand what it exactly means and what to do about it. As a security professional, I was asked by friends/family about this topic for many times. Sadly, it is usually after some major data breach happens. I believe digital privacy is a basic human right, however, companies and government did not do a good job to bring awareness to the ordinary people. Thus, I am going to start a new “personal digital privacy” series and write about simple things that everyone can understand and do to protect your digital identity and why you should absolutely do it!</p><h3>What does Google know about you?</h3><p>Ever wondered what Google knows about you? Well, if you have a Google account (like most of the people), simply follow me in the following steps to answer yourself.</p><p><strong>Step 1</strong>: Open Google map. On the righthand side bar, click on “Your Timeline”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DJquG2K3Zp4tyUlzjGFQJQ.png" /></figure><p><strong>Step 2</strong>: Next, you will be able to see all the places you have traveled to in the past few years on the map. I honestly do not remember them all, but don’t worry, Google does remember! If you click on one, it will show you a detailed description of the trip! Obviously, I took a trip in Miami 2 years ago and the record shows how much time I commute to each place! Oh, by the way, did you spot your home address and work place on the right bottom corner? From now on, when you ask your Siri to take you home, instead of feeling happy that you do not need to give him/her the full address, think twice!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HhW76OnC6SzhwQSnCFUVoQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*75QpN7y6uYFlytBEwtU_ow.png" /></figure><p><strong>Step 3</strong>: At this point, you may already try to find the delete button (but feel free to keep poking around). Click the setting button at the bottom right corner and you will be able to see several options. It is not an easy thing to do to click on that “delete all location history”, right? I know you are asking yourself: what if I want to come back to revisit these records? Another question is: even if I delete all these records, what about the records in my Gmail?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PF1ugauGTmAP_g2ZfdYkAQ.png" /></figure><h3>What can I do?</h3><p>First of all, the records may still be there even if you deleted them. However, that does not mean you should do nothing about it. A very wrong idea many people I talk to hold is that: no matter how much I do, these companies are still collecting my data and hackers are still doing bad thing. That is true, but <strong>there is no absolute security</strong> <strong>and you just need to do a little bit better than your peers! (please repeat this three times!)</strong></p><p>Let me share a fun story with you: I have to park my bike at a local train station every day and for a while it bothers me because the train station is notorious for its security issue. I find myself always worry about my bike even if I have already bought the most expensive lock and lock both the front and rear tire. Gradually, I realized, as long as there are some people park next to me use some shabby lock or simply not being careful to lock their bike, I will not be the victim. <strong>So I don’t need to be absolutely secure to keep my bike. I just need to be a little bit more secure than the people who park their bikes next to me.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/250/1*gICFSsJF4nzTRErLgMbnSg.jpeg" /><figcaption>Just be more secure than your peers!</figcaption></figure><h3>Two simple tips to be more secure than your peers!</h3><p><strong>Tip 1</strong>: Make sure you have a strong password for your Google account. Make sure you will not use it twice. Make sure it is not your birthday or whatever is guessable.</p><p><strong>Tip 2</strong>: Consider to use another email service instead of Gmail. I know this is hard because you have all your past emails/contacts in Gmail, however, information aggregation is a superpower! If Google has to know the places you have visited, don’t let it know what you are there for. Think of it in this way: if you have all your emails stored in one place and it got compromised one day, you life will be very miserable. Therefore, consider to use different email services for different purposes. I personally use Protonmail for two reasons: 1) all its servers are in Switzerland 2) it is encrypted and only supports 2FA.</p><p>It is actually simpler than you thought.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cb75c393d155" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why do I choose to get a computer science degree?]]></title>
            <link>https://lijie-zhou.medium.com/why-do-i-choose-to-get-a-computer-science-degree-9f5eeedbdc9d?source=rss-e44b1724aa04------2</link>
            <guid isPermaLink="false">https://medium.com/p/9f5eeedbdc9d</guid>
            <category><![CDATA[computer-science]]></category>
            <category><![CDATA[women-in-tech]]></category>
            <category><![CDATA[education]]></category>
            <dc:creator><![CDATA[Lijie Zhou]]></dc:creator>
            <pubDate>Fri, 19 Feb 2016 19:10:44 GMT</pubDate>
            <atom:updated>2017-09-01T05:00:34.092Z</atom:updated>
            <content:encoded><![CDATA[<p>A friend of mine at Hackbright asked about career/further study options. She comes from a biology background and is currently working as a junior C# developer. I met her at Hackbright when we both teach Hackbright’s part-time web development class.</p><p>I have been asked many times about my transition from a linguistic background into computer science. Questions people usually ask are: 1) Why would you choose a traditional computer science master’s program instead of programming boot camp? 2) Can you survive without a computer science bachelor degree? 3) What’s the return on your investment (time, money, opportunity cost)?</p><p>Q1: Why would I choose a traditional computer science master’s program instead of programming boot camp?</p><p>I am really lucky because my families (especially my boyfriend) strongly encourage me to focus on my career goal in the long term. They support me both emotionally and financially. Before I transfer to my current CS master’s program, I have already learned web development (HTML/CSS/Javascript)with <a href="https://medium.com/u/9a59e19ae2d6">Girl Develop It San Francisco</a> and some online resources. I am determined to take bigger challenges and excited for bigger rewards the technical skills are going to bring me.</p><p>Q2: Can you survive your master’s program w/o a CS background?</p><p>I have taken all the undergraduate CS classes before they allow me to get in the master’s program. I have 9 prerequisite classes and I have to get at least B for my conditional admission.</p><ul><li>CS 220 Data Structure (Java)</li><li>CS 230 Discrete Math</li><li>CS 256 Machine Structure (Intel X86 Assembly)</li><li>CS 340 Programming Methodology (C++)</li><li>CS 412 Computer Lab (Web development)</li><li>CS 413 Software Engineering (Android Development)</li><li>CS 415 Operating System</li><li>CS 510 Algorithm Analysis (Classic CS problems and runtime analysis)</li><li>CS 520 Computing Theory (Computational model and turning machine)</li></ul><p>That prepared me 3 semesters (2 years experience) before getting into my master’s program. This semester is the 1st semester I’m in the CS program as a graduate student. After passing all the basic level classes, I’m excited that I can start considering my master’s concentration. I decide to focus on cyber security in the next 2 years.</p><p>Q3: What’s return on your investment (time, money, opportunity cost)?</p><p>I think choosing a traditional CS program over a programming boot camp is a long-term game.</p><ul><li>Time: 1.5 to 2 years prerequisite if you’re not from CS background. Depending on your previous major, it may be shorter or longer. Plus, you will need to spend another two years afterwards for your master level classes. That’s the case for most of the traditional CS master’s program. Your alternative could be: programming boot camp, which usually takes 3 months; Udacity Nanodegree (about 6 months).</li><li>Money: I’m currently paying International student tuition. So if you’re a US-citizen, you may just pay one-third of my cost. The approximate cost is around $2000/credit * 60 credits = $120K. The alternatives are: Boot camp ($16K — $20K), Udacity ($200/month * 6 months)</li><li>Return on investment (ROI): my assumption is that the fundamental CS training will prepare me enough knowledge to try a lot of opportunities compared to the skill-based training. Translated into numbers, I expect to start with a backend engineer/security engineer position with the salary range from $<a href="http://www.payscale.com/research/US/Job=Security_Engineer/Salary">88,777 — $123,501</a>. Based on PayScale, one of the common career paths for security engineer is security manager, with the salary range from $<a href="http://www.payscale.com/research/US/Job=Security_Manager%2c_IT/Salary#CareerPaths">101,587 to 136,164</a>. (I use median instead of the lower bound because I believe California has higher salary bar).</li></ul><p>Thus, within 10 years time span (assuming I transfer to security management at year 7).</p><p>Total salary: 88,777*6 + 101,587*4 = $939,010.</p><p>ROI = ($939,010-$120,000)/$120,000 = 683%.</p><p>This is not a bad deal in my opinion.</p><p>I welcome any question about the career transition or my CS program.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9f5eeedbdc9d" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>