Chennai, Tamil Nadu, India
View My Portfolio: https://hardikchhipa.vercel.app/ Systems rarely break in the ways you expect them to. A dataset turns out noisier than planned, a model eats up more compute than it should, or a chip runs hotter than its datasheet promised. I enjoy living in those gaps. Tracing the root cause, fixing the bottleneck, and reworking the design until it behaves. Digging into those constraints, figuring out why things break, and designing systems that still hold up is where I do my best work. My journey has taken me across resource-constrained AI, healthcare applications, and electronics cooling research. Sometimes that means rethinking neural networks so they can diagnose on edge devices, sometimes it’s building multilingual systems that lower barriers to access, and sometimes it’s learning how to keep high-performance electronics from cooking themselves. Different fields, same principle: make technology efficient, sustainable, and usable in the real world. Along the way, I’ve published research with IEEE, ASME, and arXiv, and worked with a toolkit that spans TensorFlow, PyTorch, Hugging Face, Whisper, OpenCV, MERN, AWS, and Docker. But tools and papers aside, what drives me is curiosity, the chase of a stubborn problem until it finally gives way. And if you’ve made it this far, maybe you’re curious too. Let’s connect! who knows, the next stubborn problem might be one we solve together.
Research focusing on employing Knowledge Distillation from complex CNNs. Various CNNs (DenseNet121, DenseNet169, DenseNet201, MobileNet, ResNet152, Xception, Inception) were explored for Diabetic Retinopathy detection. Knowledge distillation was used to create a lighter student model mentored by DenseNet169, achieving 68.77% validation accuracy on a diverse Kaggle dataset, focusing on practical deployment in medical devices.