About Abhik Sarkar
What Abhik Sarkar does
Abhik Sarkar is the Director of Machine Learning at Cloudastructure Inc, where he leads a globally distributed team building computer vision systems that process over 7 million videos per day at 98% accuracy for real-time threat detection. He is known for production GPU optimization work (CUDA, TensorRT, NVDEC pipeline tuning, debugging Xid 31 MMU faults), and as a conference speaker at PyCon India, PyCon Japan, BangPypers, and DevFest on ML systems, computer vision, and Python performance.
Selected work: articles · talks · research-paper analyses · technical concepts. Browse expertise areas → See verified credentials and background →
Hey! I'm Abhik Sarkar, a seasoned Machine Learning Engineer and Full-Stack Developer with over 8 years of experience creating scalable and intelligent software solutions.
Currently, I lead the Machine Learning division at Cloudastructure Inc, where I leverage deep learning and computer vision technologies to enhance safety and security. My expertise spans diverse tech stacks including Python, C++, TensorRT, PyTorch, CUDA, Docker, and React.js, along with day-to-day work on MLOps tooling and distributed systems for large-scale video processing.
My formal training is in computer science: a Bachelor of Technology in Computer Science and Engineering from the National Institute of Technology, Raipur, followed by graduate coursework in Introduction to High-Performance Computing at the Indian Institute of Science, Bangalore, and CS224W: Machine Learning with Graphs through the Stanford Center for Professional Development.
In my leadership role, I strategically build and manage diverse global teams, hire exceptional international talent, and foster effective remote collaboration across teams distributed in 11 countries.
As a recognized speaker, I've presented technical insights at prestigious global conferences like PyCon India, PyCon Japan, and DevFests. Beyond speaking engagements, I actively engage in mentorship, workshops, and content creation, promoting knowledge sharing within the tech community.
I'm an accidental performance engineer — what started as saving the company money turned into an obsession with GPUs — squeezing every last FLOP out of CUDA kernels, optimizing memory access patterns, and understanding the hardware from TensorRT graph optimization down to streaming multiprocessor occupancy. I got hooked after realizing that decoders like NVDEC are entirely separate silicon from the CUDA cores — you can pipeline decode, preprocess, and inference across different parts of the GPU simultaneously and make the whole thing go brr. That pipelining alone gave us a 3x speedup and saved thousands in infrastructure costs at scale.
When I'm not coding or training models, you'll find me exploring new technologies, participating in quizzes, or experimenting with emerging tools and frameworks.
Follow along for insights on machine learning, AI, and software engineering through my latest publications, projects, or professional networks.
Articles I Endorse
A selection of articles that have influenced my thinking and approach to technology.