Custom AI model for agricultural pathology
Designing, training and optimizing a 152-layer model on 85,000+ images to diagnose crop diseases fast enough for real-time use.
Some questions only get answered by building. These are the threads I'm pulling right now, in AI, physics, mathematics and engineering. Each entry notes where it stands today.
Designing, training and optimizing a 152-layer model on 85,000+ images to diagnose crop diseases fast enough for real-time use.
Tokenization, stopword removal, POS tagging, NER and sentiment analysis for Bangla — open source with fully editable datasets.
Tracing how signals actually travel, from submarine cables and optical fiber to wireless communication.
Competitive problem solving in number theory, combinatorics and geometry — and the math hiding inside algorithms.
A beginner-friendly blueprint of the internet system — IP, TCP, HTTP, DNS, cloud, the dark web and more.
Field notes
I'm continuously running new explorations and documenting results. Follow along on my blog, or get in touch to collaborate on a research idea.