Hyperplane

𝗪𝗲’𝘃𝗲 𝗯𝗲𝗲𝗻 𝘁𝗵𝗶𝗻𝗸𝗶𝗻𝗴 𝗮𝗯𝗼𝘂𝘁 𝘄𝗵𝗲𝗻’𝘀 𝘁𝗵𝗲 𝗿𝗶𝗴𝗵𝘁 𝘁𝗶𝗺𝗲 𝘁𝗼 𝗱𝗼 𝘁𝗵𝗶𝘀…. While 𝗠𝗟 𝗩𝗮𝗻𝗴𝘂𝗮𝗿𝗱𝘀 reflected our early mission, it didn’t clearly show who we’re backed by or the direction we’re headed. We wanted a name that feels more aligned with our work, our identity, and 𝗼𝘂𝗿 𝗰𝗼𝗻𝗻𝗲𝗰𝘁𝗶𝗼𝗻 𝘁𝗼 Cube. And today, 𝘄𝗲’𝗿𝗲 𝗺𝗮𝗸𝗶𝗻𝗴 𝘁𝗵𝗮𝘁 𝗼𝗳𝗳𝗶𝗰𝗶𝗮𝗹. So we’re proud to introduce the new name of this newsletter: 𝗛𝘆𝗽𝗲𝗿𝗽𝗹𝗮𝗻𝗲. 𝗪𝗵𝘆 𝗛𝘆𝗽𝗲𝗿𝗽𝗹𝗮𝗻𝗲? In geometry and machine learning, a hyperplane is a decision boundary, the point where things change, shift, separate, evolve. 𝗛𝘆𝗽𝗲𝗿𝗽𝗹𝗮𝗻𝗲 𝗶𝘀 𝗼𝘂𝗿 𝘄𝗮𝘆 𝗼𝗳 𝗽𝘂𝘁𝘁𝗶𝗻𝗴 𝗮 𝗻𝗮𝗺𝗲 𝘁𝗼 Cube 𝘃𝗼𝗶𝗰𝗲𝘀, helping and making it part of something bigger. 𝗡𝗼𝘄 𝘁𝗲𝗹𝗹 𝘂𝘀, 𝘄𝗵𝗮𝘁 𝗱𝗼 𝘆𝗼𝘂 𝘁𝗵𝗶𝗻𝗸 𝗼𝗳 𝘁𝗵𝗲 𝗻𝗲𝘄 𝗻𝗮𝗺𝗲 ? #New #Hyperplane #Cube

💭 𝐖𝐡𝐚𝐭 𝐢𝐟 𝐲𝐨𝐮𝐫 𝐀𝐈 𝐞𝐱𝐩𝐞𝐫𝐢𝐦𝐞𝐧𝐭𝐬 𝐰𝐞𝐫𝐞 𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞𝐝 𝐥𝐢𝐤𝐞 𝐩𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞𝐬 𝐟𝐫𝐨𝐦 𝐝𝐚𝐲 𝐨𝐧𝐞? That’s the shift we experienced moving our deepfake detection work from 𝗖𝗼𝗹𝗮𝗯 to TractoAI. Now here’s what we noticed when we were moving: 𝗗𝗮𝘁𝗮 𝘀𝘁𝗮𝘆𝘀 𝗼𝗿𝗴𝗮𝗻𝗶𝘇𝗲𝗱 → Instead of throwing lists into a notebook, everything lives in tables. Easier to reuse, easier to scale. 𝗡𝗼 𝗺𝗼𝗿𝗲 𝗰𝗹𝘂𝘁𝘁𝗲𝗿 → Every run creates its own workspace, so inputs and outputs never get mixed up. 𝗖𝗹𝗲𝗮𝗻 𝗲𝘅𝗽𝗲𝗿𝗶𝗺𝗲𝗻𝘁𝘀 → Dependencies install fresh inside each job. If it runs once, it runs everywhere. 𝗦𝗰𝗮𝗹𝗶𝗻𝗴 𝗶𝘀 𝗯𝘂𝗶𝗹𝘁-𝗶𝗻 → Just set how much GPU, CPU, and memory you need, and let Tracto handle the rest. For us, this shift meant faster iterations, cleaner outputs, and far fewer surprises. 𝗕𝘂𝘁 𝘁𝗵𝗲 𝗿𝗲𝗮𝗹 𝘄𝗶𝗻? We’re now working in a setup that feels production-ready from day one. Don’t just take our word for it; 𝘄𝗲 𝗵𝗮𝘃𝗲 𝘁𝗵𝗲 𝗳𝘂𝗹𝗹 𝗰𝗼𝗱𝗲 𝗼𝗽𝗲𝗻 𝗼𝗻 𝗚𝗶𝘁𝗛𝘂𝗯. See how we set up the pipeline, ran the jobs, and handled results, step by step. 👀𝗟𝗶𝗻𝗸 𝗶𝗻 𝘁𝗵𝗲 𝗰𝗼𝗺𝗺𝗲𝗻𝘁𝘀 👀 #Hyperplane #GitHub #Code #TractoAI

𝗪𝗼𝗿𝗸, 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵, 𝘁𝗲𝘀𝘁𝗶𝗻𝗴, 𝗮𝗻𝗱 𝘃𝗼𝗶𝗹à ✨ It took time, iteration, and a lot of late nights, but we’re closing this project with 𝘁𝗵𝗲 𝗯𝗲𝘀𝘁 𝗿𝗲𝘀𝘂𝗹𝘁𝘀 𝘄𝗲 𝗰𝗼𝘂𝗹𝗱 𝗱𝗲𝗹𝗶𝘃𝗲𝗿. The final part of our deepfake audio detection series is 𝗻𝗼𝘄 𝗹𝗶𝘃𝗲 📢 What began as an experiment has grown into a complete system: a hybrid model trained on diverse datasets, deployed at the edge, and ready to analyze voices in real time. This last stage was only possible through our collaboration with TractoAI. Their platform gave us the environment we needed to scale beyond Colab, manage complex workflows without losing progress, and accelerate training at a level that traditional tools couldn’t match. In the end, we managed to build something that actually 𝗿𝘂𝗻𝘀 𝗼𝗻 𝗼𝘂𝗿 𝗱𝗲𝘃𝗶𝗰𝗲𝘀. No servers, no waiting, it just listens and 𝘁𝗲𝗹𝗹𝘀 𝘆𝗼𝘂 𝘀𝘁𝗿𝗮𝗶𝗴𝗵𝘁 𝗮𝘄𝗮𝘆 𝗶𝗳 𝗮 𝘃𝗼𝗶𝗰𝗲 𝘀𝗼𝘂𝗻𝗱𝘀 𝗳𝗮𝗸𝗲, whether it’s a call or an audio file someone shared. 𝗜𝘁’𝘀 𝘁𝗵𝗲 𝗲𝗻𝗱 𝗼𝗳 𝘁𝗵𝗶𝘀 𝘀𝗲𝗿𝗶𝗲𝘀, but just the beginning of where this technology can go. 𝗔𝗹𝗹 𝗽𝗮𝗿𝘁𝘀 𝗼𝗳 𝘁𝗵𝗲 𝘀𝗲𝗿𝗶𝗲𝘀, 𝗶𝗻𝗰𝗹𝘂𝗱𝗶𝗻𝗴 𝘁𝗵𝗶𝘀 𝗳𝗶𝗻𝗮𝗹 𝗼𝗻𝗲, 𝗮𝗿𝗲 𝗹𝗶𝗻𝗸𝗲𝗱 𝗶𝗻 𝘁𝗵𝗲 𝗰𝗼𝗺𝗺𝗲𝗻𝘁𝘀 👀 #AI #Hyperplane #Deepfake #Edge

Another big thanks to everyone who stuck with us, dropped feedback, or just showed some love ✨

𝗡𝗼 𝗢𝗻𝗲 𝗧𝗮𝗹𝗸𝘀 𝗔𝗯𝗼𝘂𝘁 𝗧𝗵𝗶𝘀 📱 If your assistant keeps throwing errors, maybe it’s not the architecture; it could be the missing safeguard: 𝗣𝗮𝗿𝘁 𝟯 = 𝗝𝗦𝗢𝗡 𝗖𝗛𝗘𝗖𝗞𝗦 🛡️ Here’s what you actually need to do next: Get your hands‑on guide in our “𝗕𝘂𝗶𝗹𝗱 𝗬𝗼𝘂𝗿 𝗢𝘄𝗻 𝗦𝗶𝗿𝗶” series that walks you through plugging in these checks for no more random crashes or silent failures. 👀  𝗣𝗶𝗰𝗸 𝘆𝗼𝘂𝗿 𝘄𝗲𝗮𝗽𝗼𝗻 𝗳𝗿𝗼𝗺 𝘁𝗵𝗲 𝗰𝗼𝗺𝗺𝗲𝗻𝘁𝘀 𝗯𝗲𝗹𝗼𝘄

We heard that part 3 is coming tomorrow 👀 

Nothing like a boring article to match a boring Monday  🌈 

Kind reminder 🖐

We still don't know why we love trends 💬