GDG On Campus VIT Vellore

A green light might move you forward, but it's only the outcome of countless decisions unfolding behind the scenes. A simple traffic signal becomes so much more than just a timer cycling through red, yellow, and green. It is a real-time system, constantly negotiating between competing demands: growing queues, pedestrian requests, emergency vehicle priority, incomplete information, and a future that refuses to announce itself in advance. In a world where time won't stop ticking, perfection is an impossible ask. So how do these systems function at all? Dive into how these split-second decisions are made with ‘The Art of Being Good Enough’ by Ananya Sridharan, and discover what a simple traffic signal can show us about making decisions in a world that refuses to wait. 🔗: dscv.it/rts-blog #TrafficSignals #SystemsThinking #TechBlogs

"𝘈 𝘯𝘦𝘸 𝘦𝘳𝘢 𝘰𝘧 𝘗𝘊. 25.0528, 121.5990." 𝘛𝘩𝘢𝘵 𝘸𝘢𝘴 𝘪𝘵. No poster, no trailer, no carefully worded press release. Just six words and a pair of coordinates, dropped by “NVIDIA”, “Microsoft”, and “ARM” at the exact same moment on May 29, 2026. Tech forums lit up like a city at night. Within hours, someone had punched the numbers into a map and landed in Taipei, home of Computex. The countdown had begun, and three of the world's biggest tech giants had said nothing, yet 𝘴𝘰𝘮𝘦𝘩𝘰𝘸 𝘴𝘢𝘪𝘥 𝘦𝘷𝘦𝘳𝘺𝘵𝘩𝘪𝘯𝘨. A week later, the curtain lifted, and 𝗦𝗽𝗮𝗿𝗸 stepped out of the rumour mill and onto the stage. Built by NVIDIA on ARM architecture and running Windows, it packs a 20-core processor, RTX graphics on the Blackwell architecture, and a dedicated AI engine onto a single slab of silicon. Every AI you have used so far lived somewhere else, in vast halls of humming servers, while your laptop merely knocked on their door and waited. Spark moves that intelligence onto the chip itself. Your laptop is no longer a messenger carrying requests to distant servers, 𝘪𝘵 𝘪𝘴 𝘵𝘩𝘦 𝘥𝘦𝘴𝘵𝘪𝘯𝘢𝘵𝘪𝘰𝘯. 𝗪𝗵𝗮𝘁 𝗱𝗼𝗲𝘀 𝘁𝗵𝗮𝘁 𝗰𝗵𝗮𝗻𝗴𝗲 𝗶𝗻 𝗽𝗿𝗮𝗰𝘁𝗶𝗰𝗲? Take Photoshop's generative fill, which today sends your image to Adobe's servers, processes it there, and sends it back, an internet connection and a queue standing between you and the result. On Spark, the same job runs on the chip in front of you. 𝘕𝘰 𝘶𝘱𝘭𝘰𝘢𝘥, 𝘯𝘰 𝘸𝘢𝘪𝘵, and it works on a flight as easily as at a desk. Adobe is already rebuilding Photoshop and Premiere Pro to run this way, and the same shift applies to drafting scenes from a script or cleaning up hours of footage. For thirty years, Intel and AMD have dominated this industry, powering nearly every PC on the planet. Spark is the first challenger in a generation to step into that arena, and it arrives built around AI rather than retrofitted for it. A student gets a tutor who works without a network. A founder gets an assistant who keeps working long after the wifi gives up. And because the work happens on your own machine, 𝘸𝘩𝘢𝘵 𝘺𝘰𝘶 𝘤𝘳𝘦𝘢𝘵𝘦 𝘯𝘦𝘷𝘦𝘳 𝘭𝘦𝘢𝘷𝘦𝘴 𝘺𝘰𝘶𝘳 𝘩𝘢𝘯𝘥𝘴. The first Spark laptops will hit shelves this autumn, carried by brands like “Dell”, “Lenovo”, and “ASUS”. Whether they live up to the promise, real benchmarks will decide. But the laptops of tomorrow will not simply obey, 𝘵𝘩𝘦𝘺 𝘸𝘪𝘭𝘭 𝘤𝘰𝘭𝘭𝘢𝘣𝘰𝘳𝘢𝘵𝘦. And if that era arrives, its history will not begin with a keynote. It will begin with 𝘴𝘪𝘹 𝘲𝘶𝘪𝘦𝘵 𝘸𝘰𝘳𝘥𝘴 𝘢𝘯𝘥 𝘢 𝘱𝘢𝘪𝘳 𝘰𝘧 𝘤𝘰𝘰𝘳𝘥𝘪𝘯𝘢𝘵𝘦𝘴. - Aashman Biyani

Have you ever felt like your device was holding out on you? Best believe you aren't alone. Join Shikhar Holmes on his investigation as this blog by Shikhar Sahay dives into what started as a few suspicious slowdowns and quickly spiraled into the discovery of a full-blown performance mafia. From benchmarking sting operations and hidden villains to the strange world of PC optimization and building a Windows optimization toolkit, the hunt was on to catch these little criminals red-handed. 🔗 dscv.it/optimization-blog #optimization #performance #techblogs

Have you ever wanted to stress-test humanity? Turns out if you create a fake virus and unleash it into a simulated world, you learn a lot more than expected. One superspreader can reroute the entire outbreak. A tiny mutation can rewrite the future of the network. And a single change in human behavior can decide whether the curve collapses… or civilization does. Dive into “Simulating the End of the World”, a blog by Akhyaan Kumar, which explores an epidemic simulator built using agent-based modeling, scale-free networks, stochastic probability, and SEIRD state machines to study how outbreaks evolve in real time. 🔗 : dscv.it/pandemic-blog #DataScience #Python #Simulation #Epidemiology #TechBlogs

Computers are terrifyingly smart until you realize they possess the exact emotional intelligence of a kitchen counter. They can process billions of operations a second, but the moment two different databases use the exact same word for two different things? Total, existential panic. It’s giving major "I'm fine" (but not fine) energy. The words are identical, but the backend interpretation is a Christopher Nolan level of psychological thriller. In a massive tech ecosystem, frontend devs see a "User" as a cute profile picture. Security architects see a chaotic cluster of cryptographic tokens. When thousands of decoupled microservices suffer this exact language barrier, your entire architecture enters a Bigboss level reunion standoff where everyone is screaming and nobody is listening. 𝐓𝐡𝐞 𝐌𝐨𝐧𝐨𝐥𝐢𝐭𝐡𝐢𝐜 𝐃𝐞𝐥𝐮𝐬𝐢𝐨𝐧 The immediate instinct for a junior developer is always: "𝘞𝘩𝘺 𝘯𝘰𝘵 𝘫𝘶𝘴𝘵 𝘮𝘢𝘯𝘥𝘢𝘵𝘦 𝘰𝘯𝘦 𝘨𝘪𝘢𝘯𝘵, 𝘤𝘦𝘯𝘵𝘳𝘢𝘭𝘪𝘻𝘦𝘥 𝘥𝘢𝘵𝘢𝘣𝘢𝘴𝘦 𝘵𝘢𝘣𝘭𝘦?" One actor, three teams, total chaos. When Netflix signs Pedro Pascal, everyone wants his data customized: • UI Devs demand name: { "Pedro Pascal" ; bio: "Star of The Last of Us"}; • Video Engineers need rigid tech specifics : {actor_name: "Pedro Pascal"; role: "Lead"}; • Marketing wants clout metrics:{ talent: "Pedro Pascal" ; popularity_score: 95}; This is exactly where the confusion occurs: a multi-week engineering brawl just to map three tables for one guy. . 𝐓𝐡𝐞 𝐔𝐥𝐭𝐢𝐦𝐚𝐭𝐞 𝐁𝐚𝐜𝐤𝐬𝐭𝐚𝐠𝐞 𝐒𝐡𝐚𝐩𝐞-𝐒𝐡𝐢𝐟𝐭𝐞𝐫 Netflix realized that the bottleneck wasn't a hardware limitation. It was a pure coordination problem. To solve it, they built 𝐔𝐧𝐢𝐟𝐢𝐞𝐝 𝐃𝐚𝐭𝐚 𝐀𝐫𝐜𝐡𝐢𝐭𝐞𝐜𝐭𝐮𝐫𝐞 an architectural layer driven by their internal semantic modeling engine, 𝐔𝐩𝐩𝐞𝐫. Instead of making engineers write mind-numbing , soul-crushing pipelines of manual glue code to translate between mismatched services, Netflix changed everything. Engineers now define a core business concept (like Show a or an Actor ) exactly once inside a centralized metadata Knowledge Graph. 𝐓𝐡𝐞 𝐟𝐢𝐧𝐚𝐥 𝐜𝐮𝐫𝐭𝐚𝐢𝐧 𝐝𝐫𝐨𝐩: Think of Upper as the ultimate translator. It automatically generates and projects the specific, customized technical format each team actually needs: • A clean 𝐆𝐫𝐚𝐩𝐡𝐐𝐋 𝐬𝐜𝐡𝐞𝐦𝐚 for the UI developers. • A strictly typed 𝐀𝐯𝐫𝐨 𝐬𝐜𝐡𝐞𝐦𝐚 for streaming data workers. • An 𝐀𝐩𝐚𝐜𝐡𝐞 𝐈𝐜𝐞𝐛𝐞𝐫𝐠 𝐒𝐐𝐋 𝐥𝐚𝐲𝐨𝐮𝐭 for the big data analysts. Change an attribute once at the graph level, and it propagates downstream faster than a Red Bull F1 pit stop (saving teams from sanity-draining alignment meetings). By managing their vocabulary like a flawless cinematic universe, Netflix decoupled logic from physical tables, transforming fragmented data chaos into a unified, high-speed platform. Written by Siri Reddy

Schrödinger created a cat in a box to make quantum mechanics sound ridiculous. The idea was meant to expose a flaw. If the conclusion sounded absurd enough, surely the theory itself had to be wrong. Nearly a century later, we looked at that same thought experiment and decided to engineer it into reality. This blog by Aditya Rohilla explores how superposition, entanglement, and the Cat State transformed one frustrated physicist's paradox into a real quantum circuit running on IBM hardware 1,000 times. 🔗 dscv.it/schrodinger-blog #QuantumComputing #TechBlogs #Physics

What does it take for a machine to bomb a school full of girls and call it a successful strike? Apparently, just a stale dataset and a model too confident to notice the present. The war in the Middle East has led to a development that deserves more attention and demands a critical shift in our perception of 𝘈𝘐 𝘪𝘯 𝘥𝘦𝘧𝘦𝘯𝘤𝘦. The US has been striking thousands of targets in Iran with the help of Project Maven. One particular strike that stood out and left us with a lot of lessons about how we perceive AI was the bombing of the girls' school in Iran. It was not just a humanitarian tragedy, but a case study in what happens when a machine is 𝒆𝒙𝒕𝒓𝒆𝒎𝒆𝒍𝒚 𝒄𝒐𝒏𝒇𝒊𝒅𝒆𝒏𝒕 𝒂𝒏𝒅 𝒄𝒐𝒎𝒑𝒍𝒆𝒕𝒆𝒍𝒚 𝒘𝒓𝒐𝒏𝒈. The school was located within an IRGC military compound. At some point, a wall was built separating the two, and the site was converted into a school, and that's perfectly ordinary. For a machine, however, this is the 𝘤𝘰𝘯𝘤𝘦𝘱𝘵 𝘥𝘳𝘪𝘧𝘵 𝘱𝘳𝘰𝘣𝘭𝘦𝘮. ML practitioners know this well in commercial contexts but it takes on lethal dimensions in geospatial intelligence. Military AI systems like those in Project Maven use Bayesian inference, a method that starts with a prior belief and updates it as new data arrives. You begin with "this building is a base (80% confident)" and, as more signal flows in, you update toward higher certainty. 𝐓𝐡𝐞 𝐩𝐫𝐨𝐛𝐥𝐞𝐦? The update only works if new data actually flows in. The system fails when the pipeline goes stale. Without recent satellite passes, freshly tagged imagery, or updated ground truth, the system keeps updating within its existing belief bubble. It becomes more confident without becoming more correct. Certainty and accuracy start moving in opposite directions. Imagine you were diagnosed with an illness in 2016. You come back in 2026 fully recovered, but your doctor glances at the old file. The prior is so strong that your "I feel fine" is statistically overruled. Bayesian systems are also, by design, resistant to outliers. If 97 data points say "military base" and three new ones say "school", the math will downweight those three as noise. This is usually a feature as you don't want your system to flip on a single bad data point. But when the noise is new ground truth, this becomes catastrophic. Here's the really terrifying part: statistical confidence is not the same as ground truth. A 99.9% confidence score means the model is internally consistent with its training data. It says nothing about whether the model is right about the real world. This is what temporal 𝘤𝘰𝘯𝘵𝘦𝘹𝘵 𝘧𝘢𝘪𝘭𝘶𝘳𝘦 looks like in AI. So no, the world isn't going to be undone by some dramatic AI apocalypse. It'll be undone by lazy assumptions and overfit models, by systems that confuse internal consistency with external truth. Pay attention to your priors. After all, in Bayesian inference, a bad prior is 𝘧𝘰𝘳𝘦𝘷𝘦𝘳. Unless, of course, you update it! Written by Param Vansh Singh

It’s 2:53 AM. My laptop’s dying. My eyes aren’t far behind. I open my users.json file and somehow still understand what’s going on. No asking Claude for context. Just text that explains itself. Impressive, I think to myself. Now, picture an Excel sheet storing data like: Name: Shivani, School: SCOPE Name: Neha, School: SCORE Name: Rahul, School: SENSE instead of column headers. That’s exactly what JSON looks like. Legible, friendly, and for the most part, an excel-lent deal. No pun intended. But that’s redundant, right? That’s the quiet trade-off JSON offers: 𝘩𝘶𝘮𝘢𝘯 𝘳𝘦𝘢𝘥𝘢𝘣𝘪𝘭𝘪𝘵𝘺 𝘰𝘷𝘦𝘳 𝘮𝘢𝘤𝘩𝘪𝘯𝘦 𝘦𝘧𝘧𝘪𝘤𝘪𝘦𝘯𝘤𝘺. 𝐓𝐡𝐞 𝐇𝐢𝐝𝐝𝐞𝐧 𝐂𝐨𝐬𝐭 𝐨𝐟 𝐉𝐒𝐎𝐍 JSON is essentially just plain string-based text. Easy for developers, but could be a bottleneck for machines at scale. A simple example from one of my own projects: {         "name": "Jiya",         "age": 25     },     {         "name": "Rahul",         "age": 28     } } Looks innocent. But as data grows, field names repeat thousands of times, and every object carries braces. JSON is used everywhere: APIs, browsers, configs, AI prompts, and LLM outputs. But LLMs don’t see objects. They see tokens. And here, every character matters. Serialization turns objects into strings to store: JSON.stringify(obj) Deserialization reverses it: JSON.parse(str) Each bracket, comma, and space adds characters to the token, but without any meaning. Just overhead. Extra characters = extra cost. At a small scale, it's invisible. At larger scales, it’s a tax on CPU time, memory, latency, and bandwidth. Not because JSON is faulty, but because it generously assumes a human might walk in mid-conversation and need context. But machines never do. 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐢𝐧𝐠 𝐉𝐒𝐎𝐍 𝐟𝐨𝐫 𝐋𝐋𝐌 𝐓𝐨𝐤𝐞𝐧𝐬 If machines don’t need that extra context, can we design a better format? This is where TOON (Token Oriented Object Notation) comes in. No quotes. No commas. Minimal structure. Known schema. Even if you never use TOON, this thinking leads to: • Minified JSON • Schema-driven structures • Less nesting in prompts Same data. Fewer tokens. Less punctuation expense. 𝐇𝐢𝐠𝐡 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐒𝐲𝐬𝐭𝐞𝐦𝐬 Serializing around 10,000 objects can take 200-400ms. That’s fine for small apps, but expensive for real-time systems. That’s why formats like Protocol Buffers exist. They use schemas and send compact binary messages, so field names don’t repeat with every record. 𝑺𝒐 𝒅𝒐 𝒘𝒆 𝒅𝒊𝒔𝒄𝒂𝒓𝒅 𝑱𝑺𝑶𝑵 𝒄𝒐𝒎𝒑𝒍𝒆𝒕𝒆𝒍𝒚? Not at all. JSON is still the default because it’s interoperable, debuggable, self-describing, and developer-friendly. The “hidden cost” appears only at scale, when performance matters, or tokens cost money. JSON isn’t inefficient. It’s just generous. But even generosity at scale has a cost. The real skill isn’t replacing JSON. It’s knowing when to stop refining for humans and start optimizing for machines. -Written by Shivani Patodia

3 𝐀𝐌. JPL operations room. Every monitor screaming red. Nobody touched anything. Nobody moved. Twelve of the most qualified aerospace engineers on the planet just stared. Telemetry screens flooded with temperatures that couldn't exist, voltages that defied physics. 𝐕𝐨𝐲𝐚𝐠𝐞𝐫 1, humanity's spacecraft, the most distant object our species has ever sent, was sending back gibberish. The CCSDS packet headers looked intact. The checksums passed. The data inside was pure 𝘯𝘰𝘪𝘴𝘦. SCREEN: TELEMETRY RECEIVED: 0xFFFF 0xFFFF 0xFFFF 0xFFFF EXPECTED: 0x4A08 0x00A3 0x11BC ATTITUDE STATUS: INDETERMINATE SIGNAL AGE: 22 HOURS 31 MINUTES And that's when someone said the thing nobody wanted to say out loud. "𝐓𝐡𝐢𝐬 𝐚𝐥𝐫𝐞𝐚𝐝𝐲 𝐡𝐚𝐩𝐩𝐞𝐧𝐞𝐝 𝐲𝐞𝐬𝐭𝐞𝐫𝐝𝐚𝐲." 𝘛𝘸𝘦𝘯𝘵𝘺-𝘵𝘸𝘰 𝘢𝘯𝘥 𝘢 𝘩𝘢𝘭𝘧 𝘩𝘰𝘶𝘳𝘴. That's how long Voyager's signal takes to crawl across 15 𝐛𝐢𝐥𝐥𝐢𝐨𝐧 𝐦𝐢𝐥𝐞𝐬 of vacuum. They hadn't missed an alert as 𝘵𝘩𝘦𝘳𝘦 𝘸𝘢𝘴 𝘯𝘰𝘯𝘦. By the time physics allowed them to know, the damage was already 𝘩𝘪𝘴𝘵𝘰𝘳𝘺. 𝐀 𝐜𝐡𝐞𝐜𝐤𝐬𝐮𝐦 𝐭𝐡𝐚𝐭 𝐥𝐢𝐞𝐝 𝐛𝐲 𝐭𝐞𝐥𝐥𝐢𝐧𝐠 𝐭𝐡𝐞 𝐭𝐫𝐮𝐭𝐡 A checksum passing doesn't mean data is correct. It means the transmission arrived 𝘶𝘯𝘥𝘢𝘮𝘢𝘨𝘦𝘥. The corruption was baked in at the source, a chip outputting its own memory address instead of sensor data. 𝐍𝐨𝐭 𝐚 𝐜𝐫𝐚𝐬𝐡. Just a pointer that silently rotated. As nothing validated the source, everything downstream trusted it completely. Voyager runs on an RCA 1802 chip executing 81,000 instructions per second. Your phone runs 3 billion. No operating system, no memory protection. Just human intent encoded into hardware not manufactured in 𝘧𝘰𝘳𝘵𝘺 𝘺𝘦𝘢𝘳𝘴. SCREEN: MOV R1, //SENSOR_BASE LOAD R2, [R1] NOP ; timing hold — DO NOT REMOVE NOP ; yes, both of them Nobody knew why those NOPs existed. Removing them might do nothing or kill attitude control. 𝘕𝘰𝘣𝘰𝘥𝘺 𝘥𝘢𝘳𝘦𝘥 𝘧𝘪𝘯𝘥 𝘰𝘶𝘵 𝘸𝘩𝘪𝘤𝘩. 𝐅𝐨𝐫𝐭𝐲 𝐥𝐢𝐧𝐞𝐬 𝐚𝐠𝐚𝐢𝐧𝐬𝐭 𝐟𝐨𝐫𝐭𝐲 𝐲𝐞𝐚𝐫𝐬 𝐨𝐟 𝐬𝐢𝐥𝐞𝐧𝐜𝐞 For five months they wrote forty lines, cross-referenced against paper binders from 1977, annotated by engineers who no longer exist. On 𝘈𝘱𝘳𝘪𝘭 18, 2024, they uploaded it and waited 45 hours. 𝐆𝐫𝐞𝐞𝐧. A computer built before the internet existed corrected its pointer and kept flying. It survived because someone in 1977 wrote a margin note nobody asked for, in a binder nobody opened for forty years. Thirty seconds of documentation. Forty-six years of survival. 𝐖𝐫𝐢𝐭𝐞 𝐭𝐡𝐚𝐭 𝐧𝐨𝐭𝐞. Takeaway: Checksums validate transmission integrity, not source integrity. A corrupt register will pass every downstream check. If you're not validating at the point of origin, you are trusting the pipe, not the data. That distinction doesn't matter until it's the only thing that matters. And when it does, the person debugging it won't be you. Leave them something. - Written by Sanvi Bhaisare.

36 hours of innovation, grit, and non-stop building later, WomenTechies’26 has officially crowned its champions! Every single participant impressed us with their skill, creativity and passion, but these teams were truly in a league of their own. Congratulations to all our winners for their incredible achievements and well-deserved victory! To everyone who took part, thank you for giving it your all and making WomenTechies’26 absolutely unforgettable. #WomenTechies26 #EmpowHer #BreakTheBinary