<?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 Binary Bards on Medium]]></title>
        <description><![CDATA[Stories by Binary Bards on Medium]]></description>
        <link>https://medium.com/@binarybards?source=rss-162737901924------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*QHrt3mFEX9PSAixaxVL3kg.png</url>
            <title>Stories by Binary Bards on Medium</title>
            <link>https://medium.com/@binarybards?source=rss-162737901924------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 14:48:41 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@binarybards/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[Why Ilya Sutskever Left OpenAI to Build Safe Superintelligence]]></title>
            <link>https://binarybards.medium.com/why-ilya-sutskever-left-openai-to-build-safe-superintelligence-0d36d8c1c3f1?source=rss-162737901924------2</link>
            <guid isPermaLink="false">https://medium.com/p/0d36d8c1c3f1</guid>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <dc:creator><![CDATA[Binary Bards]]></dc:creator>
            <pubDate>Fri, 18 Apr 2025 16:35:19 GMT</pubDate>
            <atom:updated>2025-04-18T16:35:19.340Z</atom:updated>
            <content:encoded><![CDATA[<p>Ilya Sutskever, a brilliant mind behind the most groundbreaking advancements in artificial intelligence, has embarked on perhaps his most ambitious journey yet. After leaving OpenAI in a dramatic split from his co-founder Sam Altman, Sutskever has launched <em>Safe Superintelligence Inc</em>. (SSI), a company with a singular focus: creating superintelligent AI that will not pose an existential threat to humanity. With significant backing and a stellar reputation, Sutskever is positioning SSI to potentially redefine how we approach AI development.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mwZKDtISpLfCgALYmhe7kg.png" /><figcaption>MIT Technology Review</figcaption></figure><p><strong>Who is Ilya Sutskever?</strong></p><p>Ilya Sutskever, an Israeli-Canadian computer scientist and machine learning pioneer, is best known as a co-founder and former Chief Scientist of OpenAI. Born on December 1986 in Russia, at just five years old, he moved with his Jewish family to Jerusalem, Israel, where they lived until he was 16, when they relocated to Canada.</p><p>He earned a Bachelor of Science in Mathematics in 2005, followed by a Master of Science in Computer Science in 2007, and culminated with a Doctor of Philosophy in Computer Science in 2013. These studies were completed under the mentorship of Geoffrey Hinton, a pioneer in neural networks whose guidance would prove instrumental in Sutskever’s development.</p><p>Sutskever’s early career moved rapidly through prestigious positions that showcased his exceptional talent:</p><ul><li>He briefly served as a postdoc researcher under Andrew Ng at Stanford University in late 2012</li><li>Co-founded DNNResearch, which was quickly acquired by Google</li><li>Worked as a research scientist at Google for nearly three years (2013–2015)</li><li>Co-founded OpenAI in 2015, where he became Chief Scientist</li></ul><p>Each step demonstrated his growing influence in the field and cemented his reputation as one of AI’s brightest minds.</p><p><strong>Contributions to Artificial Intelligence</strong></p><p>His work has fundamentally transformed how we approach machine learning and artificial intelligence, with contributions that continue to share the field today.</p><ul><li>He co-invented AlexNet with Alex Krizhevsky and Geoffrey Hinton, a breakthrough that won the ImageNet Challenge and sparked the deep learning revolution.</li><li>To support AlexNet’s massive demands, he personally purchased numerous GTX 580 GPUs online, showcasing his practical problem solving alongside theoretical brilliance.</li><li>He made groundbreaking contributions to recurrent neural networks and sequence-to-sequence learning, which is foundational to modern Natural Language Processing.</li><li>His academic work has been cited over 600,000 times within less than a decade of completing his PhD</li></ul><p>Now you understand how much he contributed to the Artificial Intelligence community and how much he meant to everyone.</p><p><strong>Why he left OpenAI?</strong></p><p>Ilya Sutskever’s scientific leadership drove OpenAI’s rise, guiding research for advanced AI systems like GPT-4. By 2023, he co-led the “Superalignment” project to ensure superintelligent AI aligns with human values.</p><p>Amid OpenAI’s success, significant philosophical differences were brewing between Ilya Sutskever and Sam Altman around topics such as how quickly advanced AI technology should be developed and how the balance between safety and commercialisation should be managed.</p><p>Around November 2023, the tech industry witnessed one of its biggest dramas with the removal of Sam Altman from OpenAI. Ilya Sutskever’s involvement in the decision as a board member sparked widespread confusion and controversy, which may have contributed to his departure from OpenAI in May 2024.</p><p><strong>The Birth of Safe Superintelligence Inc</strong></p><p>After leaving OpenAI, on June 19, 2024, Sutskever announced the formation of <em>Safe Superintelligence Inc</em>. (SSI). He co-founded the company alongside Daniel Gross, a former partner at Y Combinator and head of Apple AI, and Daniel Levy, an investor and AI researcher.</p><p>Ilya Sutskever explained, SSI would be “the world’s first straight-shot SSI lab, with one goal and one product: a safe superintelligence.”</p><p>The company established dual headquarters in Palo Alto, California, and Tel Aviv, Israel — locations deliberately chosen to leverage Sutskever’s deep connections and access to exceptional technical talent.</p><p><strong>SSI Vision and Approach.</strong></p><p>While other AI companies prioritize building capabilities first and safety second, SSI treats safety and capabilities as twin challenges to be addressed simultaneously.</p><p>This “safety-first” philosophy addresses what many AI safety experts consider a fundamental flaw in current development approaches. Rather than creating increasingly powerful systems and then attempting to constrain them afterward, SSI aims to build safety mechanisms into the foundation of superintelligent systems.</p><p><strong>Current Status and Future Trajectory</strong></p><p>Despite being less than a year old, Safe Superintelligence Inc. has already demonstrated remarkable momentum in terms of funding and partnerships.</p><ul><li>In September 2024, just three months after its founding, SSI secured an impressive $1 billion in initial funding from a consortium of venture capital firms including NFDG, Andreessen Horowitz (a16z), Sequoia Capital, DST Global, and SV Angel.</li><li>By March 2025, SSI reached a valuation of $30 billion in a funding round led by Greenoaks Capital — a sixfold increase from its previous $5 billion valuation just six months earlier.</li><li>In April 2025, Google Cloud announced a partnership to provide Tensor Processing Units (TPUs) for SSI’s research, marking a significant technical collaboration that will accelerate the company’s computational capabilities.</li></ul><p>This substantial early investments and partnerships are reflected both Sutskever’s exceptional reputation in the field and growing recognition of AI safety as a critical concern.</p><p><strong>The Significance of Safety in the AI Landscape</strong></p><p>Ilya Sutskever’s journey from OpenAI to founding Safe Superintelligence Inc. represents more than just a career change — it signals a potential inflection point in how the AI community approaches the development of increasingly powerful systems. By creating an organization solely dedicated to ensuring superintelligence can be developed safely, Sutskever has elevated the importance of AI alignment and safety to the highest priority.</p><p>The significance of SSI extends beyond its technical ambitions. It represents a fundamental philosophical stance that the pursuit of advanced AI requires dedicated focus on both safety and capabilities, not as competing priorities but as inseparable components of responsible development. This balanced approach may influence how other AI labs structure their research agendas.</p><p>In a world racing toward ever more powerful artificial intelligence, Sutskever’s vision offers a compelling alternative path — one where technological progress and human well-being advance together, ensuring that the superintelligent systems of tomorrow serve humanity’s best interests rather than endangering them.</p><p>In the rapidly evolving landscape of artificial intelligence, Sutskever’s journey from pioneering researcher to OpenAI co-founder to SSI visionary illustrates the field’s maturation — from focusing primarily on what AI can do to deeply considering what AI should do. This evolution may ultimately prove as important as any technical breakthrough in ensuring that superintelligent systems benefit humanity rather than harm it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0d36d8c1c3f1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Women in AI: The Brilliant Minds Driving the Tech Revolution]]></title>
            <link>https://binarybards.medium.com/women-in-ai-the-brilliant-minds-driving-the-tech-revolution-a74b7f8f1286?source=rss-162737901924------2</link>
            <guid isPermaLink="false">https://medium.com/p/a74b7f8f1286</guid>
            <category><![CDATA[women-in-tech]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <dc:creator><![CDATA[Binary Bards]]></dc:creator>
            <pubDate>Sun, 30 Mar 2025 18:29:36 GMT</pubDate>
            <atom:updated>2025-03-30T18:34:05.952Z</atom:updated>
            <content:encoded><![CDATA[<p>By <a href="https://www.linkedin.com/in/ashwini-radha-200061216/">Ashwini</a></p><p>The topic might sound familiar, but the story isn’t. Why haven’t women cracked the top 10 richest people in the world? Wealth aside, the real question is about impact. If you think tech mastery or intellectual edge belongs only to men, think again.</p><p>Meet five extraordinary women reshaping artificial intelligence — brilliant minds whose journeys defy the odds and redefine the future.</p><h3>Fei-Fei Li</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*fWvjZKLTgojUaYPF" /></figure><p>While many search for the ‘Father of AI,’ few ask about its ‘Mother.’ Enter Fei-Fei Li, born in Beijing, China, and now a U.S. citizen, often hailed as a <strong>Godmother of artificial intelligence</strong>. She’s the visionary behind ImageNet, the groundbreaking dataset that fueled rapid advances in computer vision during the 2010s.</p><p>At 12, her father moved from China to New Jersey; by 16, Fei-Fei and her mother followed. She earned her BA in Physics from Princeton University in 1995, working as a dishwasher during those years to support her family. Later, she completed her PhD in Electrical Engineering at Caltech in 2005, cementing her path as an AI trailblazer.</p><p>Fei-Fei Li’s career is a testament to her brilliance and vision in AI.<br><strong> → 2005–2006:</strong> She began as an Assistant Professor in the Electrical and Computer Engineering Department at the University of Illinois Urbana-Champaign.</p><p><strong>→ 2007 to 2009:</strong> She then moved to Princeton University, serving as an Assistant Professor in the Computer Science Department.</p><p><strong>→ 2009:</strong> She joined Stanford University as an Assistant Professor, rising to Associate Professor.</p><p><strong>→ 2012:</strong> Promoted from associate professor to Full professor. During this time, she directed the Stanford Artificial Intelligence Lab (SAIL) and co-founded the Human-Centered AI Institute, a university-wide initiative championing ethical AI.</p><p><strong>→ 2016</strong>: While on sabbatical from Stanford, Li joined Google Cloud as Chief Scientist of AI/ML and Vice President, focusing on democratizing AI through tools like AutoML.</p><p>In 2017, Google secured a Department of Defense contract, Project Maven, to use AI for analyzing drone footage. Despite assuring employees it was for non-offensive purposes, internal emails revealed concerns about public perception of military AI.</p><p>In 2018, Google chose not to renew the contract, with AI chief Fei-Fei Li emphasizing her stance against weaponizing AI.</p><blockquote><strong>“</strong>I believe in human-centered AI to benefit people in positive and benevolent ways. It is deeply against my principles to work on any project that I think is to weaponize AI.<strong>”</strong></blockquote><p>In late 2018, Li left Google to resume her professorship at Stanford.</p><p>Her influence didn’t stop there. In 2024, while on partial leave from Stanford, Li co-founded World Labs, a startup advancing ‘spatial intelligence’ — AI that reasons and operates in three-dimensional environments.</p><p>Today, she serves as its <strong>CEO</strong> while remaining a <strong>Stanford Professor and Co-Director</strong> of the Human-Centered AI Institute. She also co-founded AI4ALL, a nonprofit dedicated to educating diverse, next-generation AI leaders grounded in human-centered principles.</p><p>So, Do you still have a doubt why she is called as the ‘Godmother of AI’?</p><h3>Dr Lisa Su</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*1NPAAB45bR-z4M3i.jpeg" /></figure><p>Next up is Dr. Lisa Tsu-Fang Su, born in Taiwan and raised in the United States after moving there as a child. Armed with three degrees from MIT — including a PhD focused on silicon-on-insulator (SOI) technology, a then-unproven method to boost transistor efficiency — she began her career as a trailblazer.</p><p>Su was one of the first researchers to look into silicon-on-insulator (SOI) technology, a then unproven technique for increasing transistors efficiency by building them atop layers of an insulating material. Her dissertation was titled, “<em>Extreme-submicrometer silicon-on-insulator (SOI) MOSFETs</em>”<em>.</em></p><p><strong>→ 1994–1995: </strong>At Texas Instruments from 1994 to 1995, she worked in the Semiconductor Process and Device Center before joining IBM. There, as Vice President of the Semiconductor Research and Development Center, she launched an internal startup with ten employees to develop biochips and low-power semiconductors, yielding a microprocessor that extended battery life in handheld devices.</p><p><strong>→ 2007 -2009: </strong>Her ascent continued at Freescale Semiconductor, where she served as Chief Technology Officer</p><p><strong>→ 2008 -2011: </strong>Then she became a Senior Vice President and General Manager of its networking and multimedia group overseeing global strategy and engineering.</p><p><strong>→ 2012: </strong>Su joined AMD as Senior Vice President and General Manager</p><p><strong>→ 2014: </strong>Su’s appointed as President and CEO, replacing Rory Read at AMD.</p><p><strong>→ 2015–2016: </strong>She transformed AMD by shifting its focus beyond PCs, growing non-PC revenue to nearly 40% of the company’s portfolio through high-performance computing for gaming, data centers, and immersive platforms. Her leadership in FinFET-based chips and a server chip licensing deal in China drove revenue growth and a stock surge.</p><p><strong>→ 2016: </strong>Published over forty technical articles and coauthored a book chapter discussing next-generation consumer electronics.</p><p><strong>→ 2017–2019: </strong>Su spearheaded the launch of Zen-based Ryzen CPUs, including the acclaimed Threadripper line, lifting AMD’s CPU market share to nearly 11%. That same period saw her become the first woman to top the Associated Press’s CEO compensation survey, earning $58.5 million.</p><p><strong>→ 2022: </strong>She orchestrated AMD’s $49 billion acquisition of Xilinx, a leader in programmable chips, and took on the role of Chair.</p><p><strong>→ 2023: </strong>Su’s total compensation from AMD was $30.3 million, representing a CEO-to-median worker pay ratio of 238-to-1.</p><p>Today, Su is steering AMD into the AI accelerator market, challenging Nvidia with innovations like the Instinct MI300X GPU, built on chiplet technology for enhanced performance. With over 40 technical articles and a co-authored book chapter on consumer electronics, her influence spans research and industry. In 2023, her compensation reached $30.3 million, reflecting her outsized impact.</p><p>Su’s journey positions AMD as a formidable AI hardware contender — proof of her brilliance in driving tech’s future.</p><h3>Mira Murati</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*pV8egnujM9MNsd68.png" /></figure><p>Her name might be the one you anticipated in this article — and for good reason. <strong>Mira Murati, former Chief Technology Officer (CTO) of OpenAI</strong>, now leads Thinking Machines Lab, a San Francisco-based AI startup she founded in 2024. With a team of about 30 researchers and engineers — including alumni from OpenAI, Meta, and Mistral — her company aims to create advanced, understandable, and customizable AI systems for a broader audience.</p><p>Born in Vlorë, Albania, in 1988, Murati graduated from Pearson College on Vancouver Island, Canada, in 2005. She then pursued a dual-degree program in the U.S., earning a Bachelor of Arts from Colby College in 2011 and a Bachelor’s in Mechanical Engineering from Dartmouth’s Thayer School of Engineering in 2012, while interning briefly at Zodiac Aerospace.</p><p><strong>→ 2013 -2016:</strong> She joined Tesla as a product manager contributed to the development of the <strong>Model X</strong>, particularly focusing on its innovative features like the <strong>Falcon Wing doors</strong>.</p><p><strong>→ 2016–2018:</strong> She served as Vice President of Product and Engineering at Leap Motion (now Ultraleap), an augmented reality startup. Murati entered the AI spotlight, joining OpenAI as Vice President of Applied AI and Partnerships.</p><p><strong>→ 2022:</strong> Promoted to Chief technology officer (CTO) of OpenAI, she spearheaded the development of ChatGPT, DALL-E, Codex, and Sora — tools that redefined AI’s capabilities. Her work was instrumental in the development and deployment of some of OpenAI’s most notable products, such as the <strong>Generative Pretrained Transformer</strong> (GPT) series of language models. Reflecting on AI’s impact, she once remarked,</p><blockquote>“Maybe [the jobs] shouldn’t have been there in the first place”.</blockquote><p><strong>→ 2023:</strong> She briefly stepped in as OpenAI’s interim CEO after Sam Altman’s departure, returning to CTO once stability returned.</p><p><strong>→ 2024:</strong> Murati left OpenAI, expressing gratitude in her exit memo for the team’s achievements.</p><p>She soon launched <strong>Thinking Machines Lab</strong>, which, by late 2024, Bloomberg reported had reached a $9 billion valuation, with her founder stake valued at $1.4 billion — though funding details remain unconfirmed.</p><p>What sets Murati apart is her ascent: at 36, she became CTO of a booming AI giant, then boldly stepped away to build a new venture. Her determination and leadership make her a true standout in tech’s evolution.</p><h3>Joy Buolamwini</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*y2qfeUs1W4TkJRigYRhdyw.jpeg" /></figure><p>Next up is <strong>Joy Adowaa Buolamwini, a Canadian-American computer scientist and digital activist</strong> who has left an indelible mark on AI ethics. Formerly based at the MIT Media Lab, she founded the Algorithmic Justice League (AJL) to challenge bias in decision-making software, blending art, advocacy, and research to expose AI’s social implications.</p><p>Born in Edmonton, Alberta, and raised across Ghana and the U.S., she was inspired at age nine by MIT’s Kismet robot, teaching herself XHTML, JavaScript, and PHP. A competitive pole vaulter and basketball player at Cordova High School in Tennessee, she later graduated from Georgia Tech in 2012 with a computer science degree as a Stamps President’s Scholar.</p><p><strong>→ 2011:</strong> Her journey in tech began by developing an Android-based assessment system for Ethiopia’s trachoma program at the Carter Center.</p><p><strong>→ 2013:</strong> As a Fulbright fellow, she worked with Zambian computer scientists to empower youth as tech creators.</p><p><strong>→ 2016:</strong> She spoke at the White House Summit on Computer Science for All.</p><p><strong>→ 2017–2018:</strong> She earned a Master’s in Media Arts &amp; Sciences from MIT. Her Gender Shades project, supervised by Ethan Zuckerman, revealed intersectional biases in facial recognition, prompting IBM and Microsoft to refine their algorithms.</p><p><strong>→ 2020: </strong>Buolamwini’s influence soared in this year when Google and Microsoft cited her research in addressing gender and race bias.</p><p><strong>→ 2022–2023:</strong> She completed her PhD at MIT, focusing on algorithmic audits and the ‘coded gaze’ — her term for embedded biases in tech.</p><p>She advised President Biden on the AI Executive Order and published <em>Unmasking AI: My Mission to Protect What Is Human in a World of Machines</em>, highlighting AI’s social impacts and advocating for ethical AI practices.</p><p>Beyond research, she directed advocacy films like <em>Code4Rights</em> and <em>Unmasking Bias</em>, created the <em>Aspire Mirror</em>, and starred in the Netflix documentary <em>Coded Bias</em>.</p><p>Today, she leads the AJL, advising global leaders and policymakers on equitable AI while showcasing her work in exhibitions worldwide. Buolamwini’s blend of technical rigor and activism makes her a visionary in the fight for AI justice.</p><h3>Daniela Braga</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dLUAHSr9FSuJkecIRadCkw.png" /></figure><p>Dr. Daniela Braga, founder and CEO of Defined.ai, has built one of the fastest-growing scale-ups in the AI industry. Established in 2015, Defined.ai aims to provide top-tier AI data, tools, and services, creating an ethical AI marketplace that empowers customers to launch cutting-edge systems.</p><p>Braga’s journey began in Portugal, where she earned a Bachelor of Arts in Portuguese Language and Literature from Universidade do Porto (1996–2000), followed by a Master’s in Applied Linguistics from Universidade do Minho (2002–2005), with a thesis on argumentation in political debates. She then completed a PhD in Speech Technology at Universidade da Coruña (2005–2008), focusing on natural language processing for Portuguese text-to-speech systems.</p><p>→ 2008 -2012: Her career took off as a Principal Speech Scientist at <strong>VoiceBox Technologies</strong>, where she honed her expertise in AI and voice recognition</p><p>→ <strong>2012–2015: </strong>She then joined Microsoft as a Senior Speech Scientist from, advancing speech technology before founding Defined.ai.</p><p>→ <strong>2015: </strong>She<strong> f</strong>ounded Defined.ai, the company has grown into one of the leading AI data marketplaces, providing ethical AI training data.</p><p>→ <strong>2019: </strong>Awarded the João Vasconcelos Entrepreneur of the Year Award.</p><p>→ <strong>2021: </strong>U.S. President Joe Biden appointed her to a task force developing a national AI strategy. She is also a member of the World Economic Forum’s Expert Network, specializing in AI.</p><p>→ <strong>2023: </strong>Recognized by Forbes as one of the <strong>Top 10 Women Leaders in AI</strong> and among the <strong>50 Most Powerful Women in Business.</strong></p><p>Secured over <strong>$63 million</strong> in funding, including a notable Series B round, becoming the <strong>female founder</strong> to raise the largest Series B for an AI company in the U.S. Constantly seeking new challenges and pushing beyond her comfort zone, she has driven her company to remarkable success.Her story exemplifies how vision and execution set trailblazers apart.</p><p>These five women — each with unique journeys — embody the brilliance shaping AI’s future. I hope their stories inspired you as much as they did me. Stay tuned for more tech insights!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a74b7f8f1286" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Is Satya Nadella right about SaaS in the era of AI Agents?]]></title>
            <link>https://binarybards.medium.com/is-satya-nadella-right-about-saas-in-the-era-of-ai-agents-65f555f58abd?source=rss-162737901924------2</link>
            <guid isPermaLink="false">https://medium.com/p/65f555f58abd</guid>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[saas]]></category>
            <category><![CDATA[satya-nadella]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Binary Bards]]></dc:creator>
            <pubDate>Fri, 07 Feb 2025 03:56:11 GMT</pubDate>
            <atom:updated>2025-03-29T06:09:49.598Z</atom:updated>
            <content:encoded><![CDATA[<h3>AI Agents vs. SaaS: Is Nadella Right? A Tech Vet Weighs In</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UcWjmv5DKj7zxM-xlZwNsg.jpeg" /><figcaption>Credits: <a href="https://www.linkedin.com/in/rahul-dhingra/">Rahul Dhingra</a></figcaption></figure><p>By <a href="https://www.linkedin.com/in/vigneshsatya4/">Vignesh</a> &amp; <a href="https://www.linkedin.com/in/ashwini-radha-200061216/">Ashwini</a></p><p>Satya Nadella argues that most SaaS applications are essentially <strong>CRUD database with embodied business logic</strong> and rely on pre-defined workflows and user interaction</p><p>He envisions <strong>business logic shifting to an “AI tier”</strong> where intelligent agents autonomously handle cross-platform data and processes. This means AI systems could, for instance, pull customer data from a CRM, analyse financial patterns from accounting software, and generate actionable insights in real time — all without manual intervention.</p><p>His core believe is that <strong>AI agents will replace siloed SaaS workflows with unified, dynamic orchestration</strong>.</p><h4><strong>What is an AI Agent?</strong></h4><p>An autonomous software entity that performs specific tasks by perceiving environments, making decisions, and taking actions.</p><ul><li><strong>Perception:</strong> Collects data from sensors, databases, or user inputs (e.g., customer inquiries from a CRM).</li><li><strong>Decision-Making:</strong> Uses rules, optimisation algorithms, or machine learning to map inputs to outputs (e.g., classifying support tickets).</li><li><strong>Action:</strong> Executes tasks like generating responses, updating records, or triggering alerts.</li></ul><h4>Multi-Agent Systems?</h4><ul><li>Multi-Agent systems involve a network of AI agents collaborating to solve problems or achieve goals that require diverse expertise.</li><li>Imagine a team collaboration that internally communicates with each other, criticise each other, and improve each other results in order to solve a given task, vs a single agent that only get’s its feedback from the human that communicates with it.</li></ul><h4><strong>What is an AI Agentic Workflow?</strong></h4><p>A coordinated system where multiple AI agents collaborate dynamically to execute multi-step processes, adapting to feedback and changing conditions.</p><ul><li><strong>Task Decomposition:</strong> Breaks complex goals into subtasks (e.g., resolving a customer complaint into data retrieval, analysis, and resolution steps).</li><li><strong>Orchestration:</strong> Assigns tasks to specialised agents (e.g., one agent handles sentiment analysis, another routes inquiries).</li><li><strong>Tool Integration:</strong> Leverages APIs, RPA, and databases for cross-platform actions (e.g., pull customer engagement history from the Engagement Agent and updating Salesforce).</li><li><strong>Reflection &amp; Learning:</strong> Evaluates outcomes to refine future actions (e.g., adjusting response templates based on customer satisfaction scores).</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*C_CKz1E09bGZneTKYsl9LA.png" /></figure><h4>What is Vertical AI Agent?</h4><ul><li>Vertical AI Agents are specialized artificial intelligence systems designed to automate industry-specific workflows by combining deep domain expertise with autonomous task execution.</li><li>Unlike horizontal AI (e.g., ChatGPT), they focus on niche, high-impact use cases, integrating with industry tools (CRM, ERP) and proprietary datasets to deliver precision and transformative efficiency.</li><li><strong>Salient transforms the loan servicing industry </strong>with its vertical AI agents designed to streamline customer interactions and loan management processes. These AI agents operate across voice, text, email, and web chat, handling tasks like payment collection, due date changes, insurance updates, and customer queries in real-time.</li></ul><h4>How AI Agents Work?</h4><p>The modern AI agent represents a sophisticated orchestration of components working in harmony to process inputs, make decisions, and generate meaningful outputs. At its core, this architecture showcases how an AI agent processes information through a structured workflow, combining language models (LLM), memory systems, and specialised tools to deliver intelligent responses.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XaprLoWOBGyIKgqh.png" /><figcaption>Credits: <a href="https://medium.com/@craig-guanyi-li">Craig Li, Ph.D</a></figcaption></figure><ul><li><strong>Large Language Model: </strong>Brain or the decision maker of the agent.</li><li><strong>Agent Workflow:</strong> Orchestrates the data flows inside the agent.</li><li><strong>Config: </strong>Describes the agent</li><li><strong>Input and Output Proxies: </strong>They ensure that the data received and sent by the agent is in the correct format and enriched with necessary metadata.</li><li><strong>Tools</strong>: A set of functions and built-in utilities that the LLM can suggest to call to perform specific tasks. These tools can interact with external APIs, execute SQL queries, retrieve knowledge from databases, and more.</li><li><strong>Memory</strong>: Stores the prompt, context, and history of conversations. It allows the agent to maintain continuity and coherence across interactions.</li><li><strong>Knowledge</strong>: A vector database that stores information extracted from interactions and external sources, helping the agent build a comprehensive understanding over time.</li></ul><p>This architecture enables AI agents to handle complex tasks while maintaining context, accessing relevant knowledge, and producing structured outputs — all within a well-defined operational framework.</p><h4>Key Evolutionary Shifts</h4><p><strong>AI agents enable</strong> a fundamental shift from static, human-operated systems to dynamic, autonomous agents capable of complex decision-making and continuous improvement.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*T-G2E4jihIRi9FMnb1bnHQ.png" /></figure><h4>Market Landscape &amp; Key Players</h4><p>AI agents are revolutionising multiple industries by automating repetitive tasks, streamlining workflows, and improving decision-making. Several companies have already begun developing solutions across various sectors, and I have listed some of them for reference.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hmYoX8Dx_qoMIiTi7MKxXA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XM72rdh3CfI_vneQm1HxVg.png" /></figure><h4>How to build AI Agents?</h4><ul><li><strong>Scope:</strong> Identify the task, set measurable goals, and outline limitations (e.g, aim to recommend assets based on deal closure pattern, with a measurable goal of increasing conversion rates by 20% within six months).</li><li><strong>Data Pipeline:</strong> Gather relevant data, clean it, and preprocess for analysis (e.g., for predictive maintenance AI in manufacturing, collect sensor data from machines).</li><li><strong>Tools</strong>: Select beginner-friendly platforms or advanced frameworks based on expertise (e.g., LangChain, AutoGen, OpenAI Assistant API &amp; Swarm, etc).</li><li><strong>Architecture:</strong> Define components like the Agent Core, Memory Module, and Planning Module (e.g., for traffic agent, memory module to store traffic patterns, planning module to calculate optimal routes, action module to execute decisions like lane change).</li><li><strong>Algorithm:</strong> Choose suitable algorithms, train the model, and refine its performance iteratively (e.g., Use reinforcement learning (Q-learning) for a game-playing AI to optimise strategies through trial and error).</li><li><strong>Test and Debug:</strong> Perform unit, integration, and adversarial testing; debug using logs and visualisation tools (e.g., test a chatbot by automating interactions with predefined questions, using confusion matrix categories to classify responses correct, partially correct, or incorrect).</li><li><strong>Deployment and Monitoring:</strong> Deploy in the intended environment, monitor performance metrics, and update the model regularly.</li></ul><h4>How Human Agent Interfaces (HAI) will Evolve?</h4><p>As AI agents continue to rise, the Human-Agent Interface plays a vital role in shaping how humans interact with and access these systems.</p><ul><li><strong>Conversational Interfaces: </strong>Instead of manually navigating through menus or forms, users can simply type or speak requests, such as “Create a report for last month’s sales,” or “What’s the status of project X?”</li><li><strong>Proactive Agents: </strong>Instead of waiting for input, AI agents will anticipate needs and initiate actions, such as suggesting optimisations or alerting teams to anomalies.</li><li><strong>Personalised Interface: </strong>AI agents will tailor SaaS interfaces to individual users by learning from their behaviour, preferences, and needs. Instead of a one-size-fits-all approach, the platform will adapt dynamically to each user.</li><li><strong>Augmented Reality: </strong>Imagine a SaaS application for architectural design, users could view and manipulate building plans in real-time, visualizing how changes would look in their environment. AR combined with AI could create an immersive experience where users can interact with complex data or designs in more meaningful and tangible ways.</li><li><strong>Decentralised Interfaces: </strong>Imagine having a conversation in Microsoft Teams, and the AI agent assists by fetching customer data or generating reports without ever leaving the platform. This decentralisation could improve accessibility, as users wouldn’t be tied to a single platform or app to complete their tasks.</li><li><strong>UX Layout Shift: </strong>Traditional UI elements will give way to more fluid, personalised designs, seamlessly embedded within conversational canvases. This shift will streamline interactions, reducing the reliance on static menus and forms, and creating a more dynamic, intuitive experience tailored to individual user needs.</li></ul><h4><strong>How to Become the AI Agent First Company?</strong></h4><ul><li>Narrow the use case and identify where the agents work reliably (e.g, <strong>boring, repetitive tasks</strong> in industries where we can <strong>fully automate</strong> the workflow).</li><li>Build a proprietary data pipelines.</li><li>Build a strong and continuous evaluation system. (e.g, HumanEval, SWE-Bench, DS-1000, ARC, Stanford HELM)</li><li>Accelerate experimentation with new releases from OpenAI, Anthropic, Meta, DeepSeek, Qwen, or any other open-source projects.</li><li>Reach out to customers as soon as possible.</li><li>Implement a human feedback loop for continuous improvement.</li></ul><h4>Risks in Building AI Agents</h4><ul><li><strong>Cybersecurity Vulnerabilities:</strong> AI agents are prone to hacking, adversarial attacks, and data breaches, compromising security and privacy.</li><li><strong>Bias and Discrimination:</strong> Biased training data can lead to unfair or discriminatory outcomes in decision-making.</li><li><strong>Lack of Transparency:</strong> Black-box systems make it difficult to understand or trust AI decisions in critical applications.</li><li><strong>Social Risks:</strong> Excessive dependence on AI can reduce human decision-making and foster social isolation.</li><li><strong>Ethical Dilemmas:</strong> Autonomous decisions raise accountability and fairness concerns in morally complex scenarios.</li></ul><h4>Challenges in Developing AI Agents</h4><ul><li><strong>Data Quality and Availability: </strong>Poor or insufficient data leads to ineffective and biased AI models.</li><li><strong>Algorithm Selection and Training:</strong> Choosing and training the right algorithms requires expertise and significant resources.</li><li><strong>Integration with Existing Systems:</strong> Ensuring compatibility with current workflows is often complex and resource-intensive.</li><li><strong>Regulatory Compliance: </strong>Adhering to laws on privacy, ethics, and accountability adds significant complexity to development.</li><li><strong>Trust and User Control:</strong> Users may distrust AI due to a lack of control over its tools or decision-making processes.</li></ul><h4>Where are AI agents headed?</h4><ul><li><strong>Superagents</strong> represent the next evolution of AI systems, designed to autonomously handle complex, goal-oriented tasks with minimal human intervention.</li><li>Superagents<strong> </strong>go beyond traditional AI agents by integrating advanced reasoning, adaptability, and decision-making capabilities, enabling them to perform tasks typically requiring PhD-level expertise.</li></ul><h4><strong>Conclusion</strong></h4><ul><li>While AI agents are undeniably disruptive, <strong>they are more likely to transform the SaaS landscape than eliminate it entirely.</strong> The future will likely see a convergence of traditional SaaS and Agentic AI systems, where both coexist and complement each other.</li><li>A compelling example is how Salesforce CRM provides core features like data storage, user interfaces, and workflow management, while Einstein Copilot acts as an intelligent overlay that automates repetitive tasks, enhances decision-making through predictive analytics, and personalises user interactions in real-time.</li><li>Hopefully, Satya Nadella does not literally believe that SaaS is dead, but he is arguing that a transformative shift in the SaaS model will happen due to the rise of AI agents.</li></ul><p>Thanks for reading 😊</p><p><strong>References:</strong></p><ol><li><a href="https://aimresearch.co/ai-startups/vertical-ai-agents-will-dominate-2025">https://aimresearch.co/ai-startups/vertical-ai-agents-will-dominate-2025</a></li></ol><p>2. <a href="https://medium.com/binome/designing-llm-based-agents-key-principles-part-1-7e8c6fe3ddaf">https://medium.com/binome/designing-llm-based-agents-key-principles-part-1-7e8c6fe3ddaf</a></p><p>3. <a href="https://medium.com/@elisowski/ai-agents-vs-agentic-ai-whats-the-difference-and-why-does-it-matter-03159ee8c2b4">https://medium.com/@elisowski/ai-agents-vs-agentic-ai-whats-the-difference-and-why-does-it-matter-03159ee8c2b4</a></p><p>4. <a href="https://medium.com/@oriziv4/how-ai-agents-will-disrupt-saas-in-2025-7567d793ca68">https://medium.com/@oriziv4/how-ai-agents-will-disrupt-saas-in-2025-7567d793ca68</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=65f555f58abd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Pivoting Dreams: How I Found Engineering Beyond Medicine]]></title>
            <link>https://binarybards.medium.com/pivoting-dreams-how-i-found-engineering-beyond-medicine-d98a922558e0?source=rss-162737901924------2</link>
            <guid isPermaLink="false">https://medium.com/p/d98a922558e0</guid>
            <category><![CDATA[dreams]]></category>
            <category><![CDATA[women-in-tech]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[engineering]]></category>
            <dc:creator><![CDATA[Binary Bards]]></dc:creator>
            <pubDate>Wed, 01 Jan 2025 16:04:13 GMT</pubDate>
            <atom:updated>2025-01-01T16:04:13.991Z</atom:updated>
            <content:encoded><![CDATA[<p>By <a href="https://www.linkedin.com/in/ashwini-radha-200061216/">Ashwini</a></p><p>Most Indian kids’ dream is to become a doctor; I’m not very different from that. As I grow, I’m not very sure about what I’m going to do with my dreams and the possible ways to achieve them.</p><p>During my higher education, I chose the Biology-Mathematics stream, not because I had a clear vision of becoming a doctor, but due to the influence of friends and family members who suggested it was a flexible choice for future career options in Medicine and Engineering.</p><p>The study of human anatomy fascinated me, and I felt a strong desire to pursue medicine. Despite my love for biology, the road to medicine was filled with challenges — particularly the NEET exam and financial constraints. My eligibility score for Medicine was higher than for Engineering. Considering everything, I reluctantly decided to pursue Engineering instead.</p><p>I love computers, and I grew my interest and knowledge in programming during summer holidays. Due to my interest in computers and technology, I decided to study Computer Science or Information Technology, and I believed this would help me acquire knowledge about both. Because of my strong academic performance in school, I qualified for a prestigious scholarship offered by an NGO. I was thrilled when I secured admission to a well-regarded university.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*TlQ5D3a6R4bLWMwS.jpeg" /></figure><p>However, things didn’t go as planned — I couldn’t get into the Computer Science or Information Technology stream I initially wanted and ended up in Electrical and Electronics Engineering. This wasn’t ideal, as only two courses in the curriculum were related to programming. Fortunately, I made the most of my electives by choosing courses in Machine Learning and Artificial Intelligence.</p><p>The COVID-19 pandemic further limited opportunities for extracurricular activities and tech events, but I kept pushing forward. In my pre-final year, I was selected as a Software Engineer at a top-tier multinational company during campus placements. Everything seemed to be falling into place. Yet, challenges resurfaced. I wasn’t allowed to attend further interviews for higher-paying roles since I had already been placed. Then, I received a shocking email from the company I was selected by, stating that five students — including me — needed to take a few more tests, but with no further details. Despite contacting the organization, we received no reassurance. It felt like walking on a tightrope, not knowing whether I would be employed at all. The sad part is that I was not allowed to attend other interviews since I had already been selected. This felt like a significant blow.</p><p>During this uncertain period, I received a call from a mentor at the NGO who offered me an internship at a fast-growing tech firm. The prerequisites for the internship were HTML, CSS, JavaScript, React JS, and Java. However, the only programming language I was familiar with was Python learned while doing my elective courses. To add to my anxiety, all the other interns were from computer science background, and one was especially tech-savvy. Despite feeling overwhelmed, I resolved to work hard and not back down.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ZZXKH5BuAwVrqk1S.jpeg" /></figure><p>I started learning everything from scratch, created web pages, added styles, and gradually moved on to working with APIs. I completed small projects, built a chat application, and even delivered a presentation on Blockchain — all of which added value to my profile.</p><p>After three months, I faced a tough decision. The multinational company I had been selected for earlier finally called me for a pre-onboarding meeting, while the tech startup where I interned offered me a full-time role. After much deliberation, I chose to stay with the startup, valuing the opportunity for growth and learning in a dynamic environment.</p><p>Now, after nearly two years with this company, I’ve grown both professionally and personally, and I was honoured to receive the Best Employee award in April 2024.</p><p>I realized that life does not always give us what we wish for, but it gives us the freedom to make the right decisions and realize what we’re really good at and what we want to become.</p><p>I thought about starting my writing journey with tech articles, but I wanted to begin by sharing the story of how I got started. I’ll write about engineering and technology in my upcoming articles. Thanks.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d98a922558e0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Auto Scaling Delivers Personalised IPL Viewing Experiences: CSK vs MI]]></title>
            <link>https://binarybards.medium.com/how-auto-scaling-delivers-personalised-ipl-viewing-experiences-csk-vs-mi-8f5f47fb7164?source=rss-162737901924------2</link>
            <guid isPermaLink="false">https://medium.com/p/8f5f47fb7164</guid>
            <category><![CDATA[performance]]></category>
            <category><![CDATA[scalability]]></category>
            <category><![CDATA[saas]]></category>
            <category><![CDATA[autoscaling]]></category>
            <category><![CDATA[platform-engineering]]></category>
            <dc:creator><![CDATA[Binary Bards]]></dc:creator>
            <pubDate>Sun, 06 Oct 2024 12:50:01 GMT</pubDate>
            <atom:updated>2024-11-04T02:01:48.583Z</atom:updated>
            <content:encoded><![CDATA[<h3>How Auto Scaling Enhances the Delivery of Personalized IPL Viewing Experiences for the Audience</h3><p>By <a href="https://www.linkedin.com/in/vigneshsatya4/">Vignesh</a> and <a href="https://www.linkedin.com/in/ashwini-radha-200061216/">Ashwini</a></p><p>Let’s imagine we are live streaming the first match of IPL 2025 between Chennai Super Kings and Mumbai Indians.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0EroJzkDm9Rd5rOPYX1umg.jpeg" /></figure><p>Thanks to the BCCI’s uncapped player policy, MS Dhoni is making a surprising return to play this season, generating a wave of excitement among fans. Alongside him, Hardik Pandya leads the Mumbai Indians, and millions are tuning in to catch the action.</p><p>Dhoni’s comeback has fans eager to see if he can still deliver his signature finishing style, while Mumbai supporters are focused on Rohit’s explosive batting and Jasprit Bumrah’s exceptional bowling. A one-size-fits-all viewing experience won’t satisfy everyone, as each fan has their own unique preferences and interests in the game.</p><p>We are using machine learning algorithms on AWS EC2 instances to analyze viewers’ emotional reactions in real time during the match between Chennai Super Kings and Mumbai Indians, as well as their past viewing preferences and favorite players. This system can automatically compile and deliver highlight moments to viewers shortly after they happen, significantly enhancing their overall experience of the game.</p><blockquote>In this article, we will focus on auto-scaling rather than the algorithms that create personalised content for viewers.</blockquote><p>As more viewers tune in to watch a match, analysing their data and creating personalised viewing experiences demands significant resources. Auto-scaling helps us handle this increased traffic by automatically allocating additional resources when there are many viewers, ensuring that personalized content is delivered to everyone. This approach allows us to provide an excellent viewing experience during peak times while reducing costs when fewer people are watching, all without the need for constant manual adjustments.</p><p>I hope you have a basic understanding of how auto scaling works and its importance. Now, let’s explore how to set it up in AWS.</p><p><strong>We need to setup the following components to handle traffic and setup additional servers.</strong></p><ul><li>An alarm that triggers scaling policies</li><li>Launch template with Amazon Machine Image (AMI)</li><li>Auto Scaling Groups and Scaling Policies</li><li>Streamlined termination of instances</li></ul><h4>Configure an Alarm to Trigger Scaling Policies</h4><p>We need a metric to set up an alarm that will trigger the scaling policies; in our case, this metric is the viewer count. We are running a Python script that retrieves the number of viewers every ten seconds and pushes this data to CloudWatch metrics.</p><pre>def push_metric(value):<br>    response = cloudwatch.put_metric_data(<br>        Namespace=&#39;ContentGenerationMetrics&#39;,<br>        MetricData=[<br>            {<br>                &#39;MetricName&#39;: &#39;ViewersCount&#39;,<br>                &#39;Value&#39;: value,<br>                &#39;Unit&#39;: &#39;Count&#39;,<br>                &#39;Timestamp&#39;: time.time(),<br>            },<br>        ]<br>    )<br><br>    print(f&quot;Metric pushed: ViewersCount with value {value}&quot;)</pre><p>The alarm will activate scaling policies if the viewer count exceeds 90,000 for a duration of 10 seconds.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*P7y9RAbJd-6BQXPZfDFnmg.png" /></figure><h4><strong>Setup a launch template with Amazon Machine Image</strong></h4><p>I have already created an Amazon Machine Image (AMI) from my existing EC2 instance, where I configured all the necessary dependencies and applications for processing user data and generate content. Please refer to an <a href="https://medium.com/twintech/deploy-aws-instance-with-pre-installed-softwares-using-amazon-machine-image-ami-b8a040d8029f">article</a> that explains what an AMI is and how to create one.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Qc5TUtagfQenyYIvQ7gYlA.png" /></figure><p>Create a launch template with the AMI, making sure to choose the right image, instance type, suitable security policies, and other configurations. This will ensure that the application is launched with the correct settings and permissions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tvugXNoa-uNvz8aSfMLJIQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*T8CBBdlhAS99ZDb4725ciQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xHlwAoyxCdtnyp94uxpbtg.png" /></figure><p>We are now prepared to set up the auto-scaling group and scaling policies.</p><h4><strong>Define Auto Scaling Groups and Scaling Policies for Capacity Adjustment</strong></h4><p>To create an auto-scaling group, it is necessary to choose a launch template, which will assist in launching new instances when there is a higher load.</p><p>After selecting the launch template, you can configure the network, desired capacity, minimum and maximum capacity limits, instance maintenance policy, scale-in protection, and other settings.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*y-aQbgxurwdCr48G6hyCXA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xeRraAnl9C9ytML2vqIZ5A.png" /></figure><p>Once the auto-scaling group is created, we can define scaling policies for adding or removing instances.</p><ul><li><strong>Target Tracking Policy:</strong> This policy automatically adjusts the number of instances to maintain a specified target metric, such as keeping CPU utilisation at 50%.</li><li><strong>Step Scaling Policy:</strong> This policy adds or removes instances in response to specific metric thresholds. For instance, if the number of users increases by 100,000, it will add one instance for every additional 10,000 users.</li><li><strong>Simple Scaling Policy:</strong> This policy adjusts the number of instances based on a single metric threshold, allowing for straightforward scaling decisions.</li></ul><p>We are currently utilising 10 m5.8xlarge EC2 instances to analyse user behaviour in real-time and generate content dynamically. Each instance can support up to 10,000 users, and we plan to add an additional instance for every 10,000 new users.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NsQWOTPFEZQtE9tTCntIeQ.png" /></figure><p>In South India, our platform has 200,000 paid users and an average of 125,000 IPL viewers. We do not offer a customized viewing experience for free users, who can only watch a live match for 10 minutes on match day. Consequently, we have implemented an autoscaling policy to support up to 160,000 users in real time.</p><h4>Streamlined termination of instances</h4><p>It’s crucial to manage the instance termination process properly, as the scale-in policy may terminate an instance without considering what is running inside, which can lead to abrupt stoppages of ongoing processes.</p><p>A lifecycle hook allows us to postpone the termination process for a specified period, enabling ongoing processes within the instances to finish and ensuring the lifecycle is completed properly.</p><p>We can also set a default result if the lifecycle completion fails. In our case, we’ve instructed it to abandon the termination, as it’s essential to provide the best real-time viewing experience.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bln0PCdSEkeW0zqscIS3Zg.png" /></figure><p>Using SNS, we can set up a REST API endpoint to receive lifecycle hook notifications and complete the process once the ongoing tasks are finished.</p><pre><br>def check_process_complete():<br>    # Logic to check if there are any ongoing tasks or if the process is complete<br>    # Return True if the process can terminate, False otherwise<br>    return False  # Replace with actual condition<br><br>def can_terminate():<br>    for i in range(10): <br>        if check_process_complete():<br>            print(f&quot;Process complete after {i+1} checks.&quot;)<br>            return True<br>        else:<br>            print(f&quot;Check {i+1}: Process not complete. Waiting 1 minute...&quot;)<br>            time.sleep(60) <br><br>    print(&quot;Process did not complete in 10 checks.&quot;)<br>    return False<br><br>@app.route(&#39;/sns-notification&#39;, methods=[&#39;POST&#39;])<br>def sns_notification():<br>    message = json.loads(request.data)<br>    <br>    # handle subscription confirmation<br>    if message.get(&#39;Type&#39;) == &#39;SubscriptionConfirmation&#39;:<br>        confirmation_url = message.get(&#39;SubscribeURL&#39;)<br>        return jsonify({&quot;status&quot;: &quot;Subscription confirmation received&quot;}), 200<br>    <br>    # check if process can be terminated<br>    if not can_terminate():<br>        return jsonify({&quot;status&quot;: &quot;Process could not be terminated after 10 attempts&quot;}), 408<br><br>    # assume this is a lifecycle hook notification<br>    lifecycle_hook_name = message.get(&#39;LifecycleHookName&#39;)<br>    auto_scaling_group_name = message.get(&#39;AutoScalingGroupName&#39;)<br>    lifecycle_action_token = message.get(&#39;LifecycleActionToken&#39;)<br>    lifecycle_transition = message.get(&#39;LifecycleTransition&#39;)<br>    <br>    try:<br>        # complete lifecycle action<br>        autoscaling = boto3.client(&#39;autoscaling&#39;)<br>        autoscaling.complete_lifecycle_action(<br>            LifecycleHookName=lifecycle_hook_name,<br>            AutoScalingGroupName=auto_scaling_group_name,<br>            LifecycleActionToken=lifecycle_action_token,<br>            LifecycleActionResult=&#39;CONTINUE&#39;<br>        )<br><br>        return jsonify({&quot;status&quot;: f&quot;Lifecycle action {lifecycle_hook_name} completed successfully for {lifecycle_transition}&quot;}), 200<br><br>    except Exception as e:<br>        return jsonify({&quot;error&quot;: str(e)}), 500</pre><p>With this setup we can able to deliver tailored viewing experience to our viewers.</p><p>Thanks for reading</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8f5f47fb7164" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>