Sanchit Narula’s Post

As #Linkedin keeps since few days pestering me with questions about how AI improved something that I do ... ... look at th post on Facebook 😂my own private multi-team concurrent hackatons Experiment that repeated few times, and reduced my pile of backlog concepts to explore while increasing reuse Only critical component: gradually, became more proficient and more structured But hit the paywall barrier more often (I use the free versions in these initial phases to avoid surprises, only what passes the filter but require more will be eventually carefully used on paid versions) Reason? Older prompts got lame and shorter versions,and took few rounds to extract answers Now that probably both the models and (mainly) I improved our interactions, usually models hit the ground running with "gusto", and get forward toward the stated aim under the provided conditions often in just one step https://lnkd.in/dxaPnUUK Disclosure: between since the late 1980s I was few times in few countries involved in those software selections for business users where actually vendors were asked to carry out similar activities- hence, I was not re-inventing the wheel, just adapting to a different context...

🚨 Have you ever pondered the mechanics behind Facebook's lightning-fast delivery of billions of notifications? In a recent interview, I was challenged with the task of designing Facebook's Notification System. The seemingly straightforward question unravels complexities when considering real-time delivery, personalization, deduplication, and handling massive scale. My proposed solution entails an event pipeline, a notification service tailored to preferences, and a robust delivery layer encompassing push notifications, in-app alerts, and emails, all while navigating the intricacies of trade-offs. The key insight here is that the pursuit of the "perfect design" gives way to the importance of dissecting problems and strategizing for scalability. How would you tackle this challenge? #SystemDesign #ScalableSystems #BackendEngineering #InterviewPrep #SoftwareEngineering #DistributedSystems #TechInterviews #CareerGrowth #BigData #CloudArchitecture

Discover how Facebook’s FOQS (Facebook Ordered Queueing Service) effectively manages trillions of items daily, powering notifications, video tasks, and more! 🌐 ⚡ Scalability: FOQS seamlessly handles vast queue operations across numerous services, ensuring efficiency and reliability. 🗃️ Organization: Work is meticulously organized by namespaces and priority queues, optimizing processing for various tasks. 🔑 Multi-tenancy: Each team enjoys isolation in processing workflows, preventing congestion and enhancing performance. ⚙️ From language translation to backend batch processing, FOQS exemplifies the power of well-designed distributed priority queues and relational storage at massive scale. Explore more about this innovative system here:(https://lnkd.in/g_uTyR9d) #DistributedSystems #Scalability #MySQL #SystemDesign #Facebook #Queues #TechInfrastructure #BigData #Backend 📈📊

Every time you refresh your feed, different posts appear. Not because of algorithms. Because distributed systems can't agree on what's real right now. Welcome to eventual consistency. Your Instagram likes might show 47 on mobile and 52 on desktop. Both are "correct" at that moment. The system will sync eventually. 🔄 Performance vs Perfect Accuracy •NoSQL databases like Cassandra prioritize speed •Data spreads across multiple servers globally •Updates take time to propagate everywhere •Users get faster responses, slightly stale data When to embrace eventual consistency: •Social media feeds (Facebook, Twitter) •E-commerce product catalogs •Content delivery networks •Real-time analytics dashboards When to avoid it: •Banking transactions •Inventory management for limited stock •Authentication systems •Payment processing Recent developments are pushing boundaries: •CRDTs (Conflict-free Replicated Data Types) make merging easier •Edge computing brings data closer to users •New consensus algorithms reduce sync time The reality? Most apps you use daily run on eventual consistency. Your Netflix recommendations, Spotify playlists, even LinkedIn connections - they're all "eventually" correct. The internet feels instant because we've accepted "good enough, right now" over "perfect, but slower." What's your experience with data consistency challenges in your projects? #DistributedSystems #BackendDevelopment #SoftwareEngineering

🚗 Facebook Marketplace scraping (again) Sounds easy: “Just bring me car listings into Telegram.” But here’s the reality: - Facebook keeps changing its logic and filters - Accounts get blocked if not properly purchased and warmed - Old code vs. tweaked versions must be compared carefully - Even with filters, junk results will appear That’s why these projects are never “just a scraper.” It’s always about keeping it stable and running tomorrow, not only today. Our approach: - Compare code with previous commits to see what changed - Clarify filters, frequency, and data flow - Stabilize accounts & proxies (our “special sauce”) - Propose support, so client avoids the “run → break → fix” loop 💡 The value? Not in pushing code once. But in making sure it doesn’t collapse after the next Facebook update. In reality, when Facebook changes, automation has to change as well.

"Reddit lost $6.5b in market cap" - The latest development is more about why and how this could even happen. You remember when they said LLMs don't rely on top 10 search results, but on long-tail queries, generating answers from across the web? Well that web became inaccessible by ChatGPT as Google removed of its 100 search pages by simply disabling &NUM=100 and it became difficult for ChatGPT to source its information.. But why did it in impact Reddit? Reddit had millions of conversations ranking in those after 10 search pages, which wiped off its visibility as they became inaccessible. So ChatGPT didn't stop citing Reddit, it happened as a result of what Google did.. BTW Google stocks are 🚀

What is causing the disruption to Reddit? Not a scandal. Not a bad quarter. Just a single API change from Google. Google quietly replace the "num=100" parameter to "num=10" in its Search API. That small setting once allowed developers to pull the top 100 search results. Now it’s capped at 10. Sounds harmless? It’s not. That change has erased almost 90% of the internet’s discoverability for AI tools, SEO systems, and content aggregators that rely on Google’s data. Everything beyond page one, positions 11 to 100, is now invisible to machines. And that’s where Reddit feeds live. Millions of community discussions, niche topics, and early ideas, all in the “long tail” of the web. When that layer went dark, Reddit’s visibility dropped. Investors noticed. Its stock price reportedly fell more than 15% over a month This is the kind of disruption we rarely talk about. Not the kind that comes from a startup. But the kind that comes quietly when infrastructure rules change. When discovery breaks, entire ecosystems shake. Because in today’s world, whoever controls visibility controls innovation. That’s the real disruption story behind Reddit’s dip and a warning for every platform that depends on borrowed discovery. #Innovation #AI #Disruption #Google #Search #APIs #DigitalStrategy #Technology #FutureOfTech #ProductManagement #DataEconomy

I just wrapped up one of my most frustrating (yet mildly rewarding) projects yet. My mom runs an e-commerce business and spends hours every day answering the same customer support question: “Where can I find product information for X?” Her answer is always the same: “On my website!” Naturally, I saw this as a perfect chance to flex my automation skills. The plan sounded simple enough: 1.) Connect to the Facebook API 2.) Run comments through a sentiment-analysis agent 3.) Reply with the right link Easy, right? Wrong. Turns out Facebook’s API is more of a boss battle than a tool—at one point I discovered you might actually need a business license just to get your app approved. After many late nights, countless dead ends, and a suspicious amount of reliance on Google’s AI assistant, the workflow is finally live (though very much in beta). If anyone here has successfully wrangled the Facebook API for a similar project, I’d love to hear your tips—especially when it comes to debugging.

Look at that red line in the picture attached. That's Reddit disappearing from the internet in 48 hours. Through early September, ChatGPT mentioned Reddit in 10-14% of its answers. Then on September 10th, it dropped to under 2%. An 85% collapse overnight. What happened? Google removed a technical setting called "num=100" that let systems view 100 search results at once instead of 10. ChatGPT relies on Google's search results. When access dropped from 100 to 10, Reddit vanished. Most of their content ranked between positions 11-100. Outside the top 10, you are invisible. due to this Reddit's stock fell from $240 to $205 in five days. $6.5 billion gone. Not because their content changed, but because the access changed. Meanwhile, LinkedIn, YouTube, and Medium stayed flat near zero. They weren't heavily featured in ChatGPT responses to begin with. At least they're still visible in Google's top 10. This goes beyond one company's stock price. CNN is down 30% in traffic. Business Insider and HuffPost? Down 40%. News sites collectively lost 600 million monthly visits in twelve months. Google's "zero-click searches" jumped from 56% to 69% in a year. Seven out of ten searches now end without anyone clicking a website. People get their answer directly from Google. The sites that created the content get nothing. For small businesses, this is quietly devastating. You used to rank on page 2 or 3 and still get customers. Now if you are not in the top 10, you don't exist. More people ask ChatGPT instead of Googling. Invisible to AI means invisible to customers. The internet didn't shrink. The visible internet did. From 100 positions that mattered to 10. This is about power. Google and OpenAI control what gets seen. One parameter change and companies vanish. Reddit had million-dollar licensing deals. They thought they were essential. Then one switch flipped, and billions in value evaporated. Most people using these tools don't know this happened. They are getting answers, making decisions. They don't see what's missing. They don't know the sites helping them are dying because traffic disappeared. You can build something great, but if you are outside the top 10 that AI sees, you might as well not exist. The internet just became winner-takes-all.

Everyone talks about Gemini and GPT, but hardly anyone is exploring Grok 4 Most people see Twitter (X) as just social media – but for real-time research, it’s a goldmine. Google shows you what’s already SEO-optimised. Twitter shows you what’s happening right now. I use these 3 Grok-powered prompts daily for my classroom teaching (Twitter-first, not Google): I’m using these prompts daily with Grok 4. You should try too—your research game will completely change for your classroom teaching Prompt: Search X and the web for the latest updates on [specific topic, e.g., AI regulations; Tesla stock] as of [March 17, 2025]. Filter results by [preference, e.g., most credible, most recent, from experts], and highlight [focus, e.g., key events, opinions, stats] in a brief summary. Prompt: Evidence Checker Check this claim: [specific statement, e.g., ‘Eating garlic boosts immunity by 50%.’] Search for evidence from [source type, e.g., studies, X posts, news] and tell me if it’s [criteria, e.g., well-supported, dubious, mixed], including [details, e.g., sample size, date of research]. Prompt: Reliable Source Hunter Find me [number, e.g., 3] reliable sources on [subject, e.g., home solar panels] from [platform, e.g., web, X both]. Ensure they’re [criteria e.g., recent within 6 months, from experts, data-driven], and summarize [focus, e.g., cost, pros/cons] for each.