FilterX
Digital Rakshak
The problem FilterX solves
FilterX protects users from accidental or intentional exposure to adult/NSFW content online, something traditional blockers fail to do reliably. It detects and blocks harmful images, videos, slang, and URLs in real time, even when they appear inside thumbnails, social media feeds, or Indian-language text. This makes digital spaces safer for students, schools, hostels, workplaces, and public Wi-Fi networks, who struggle to monitor content misuse. Our solution removes distractions, prevents early exposure, and creates a cleaner, safer browsing experience without slowing down the system or compromising privacy.
Challenges we ran into
One major challenge was handling inconsistent and unbalanced NSFW datasets, which caused the model to overfit and misclassify borderline-safe images. We solved this by reorganizing the dataset, applying heavy augmentation, and fine-tuning MobileNetV2 with better metrics like precision/recall. Another hurdle was running real-time inference inside a Chrome Extension, since TFLite files are large and browser execution is restricted. We overcame this by converting the model to TensorFlow Lite, reducing size, and optimizing the client-side code to load the model asynchronously. Lastly, syncing data between Colab, Kaggle, and GitHub frequently corrupted ZIPs, so we switched to a clean pipeline and fixed folder structure.
Technologies used