I'm an AI Scientist with a decade of experience across smart devices, healthcare, and enterprise AI, currently leading AI, ML, and Data Science teams delivering scalable GenAI applications and robust ML systems serving millions of users. The best way I've found to truly understand something is to derive the math, write the code, and dig into every layer: no black boxes, no hand-waving. Everything I learn becomes a public artifact, whether it's a derivation, an annotated implementation, a paper breakdown, or an open-source tool.
- Lead & Ship: Guide cross-functional teams in developing production-ready AI/ML and GenAI systems serving millions of users across multiple industries
- Derive & Build: Work through ML's mathematical foundations from first principles and rebuild the machinery of modern deep learning from scratch (GPT-2, LSTMs, transformers, a PyTorch-like framework), with every line annotated with the reasoning behind it
- Write & Share: Author long-form posts, mathematical deep dives, and structured paper breakdowns on my blog. Writing is how the understanding forms, not just how it gets reported
- Contribute: Actively open-source what I build, believing that knowledge should be freely accessible
- Mentor & Teach: Develop talent within my teams and the broader community. Teaching is how the learning happens
- Apply: Distill a decade of battle-tested lessons on what actually works when models leave the notebook and hit production
- 🤖 Building scalable and reliable GenAI applications for massive user bases
- 📐 Rebuilding the mathematical anatomy of learning from the ground up: probability, optimization, information theory, and learning theory
- 🔬 From-scratch, annotated implementations of the systems behind modern deep learning
- 📄 Structured breakdowns of foundational and influential papers, distilled into what matters and why
- 🛠️ Open-source tools that accelerate ML experimentation
Lift the hood on deep learning. Build Tensors, autograd, and CUDA backends from scratch, because you don't understand it until you can rebuild it.
📦 cmn_ai
Most ML projects fail slowly… because iteration is slow.
cmn_ai kills the boilerplate so you can go from idea → experiment → result in hours, not weeks.
- 🧠 Attention Is All You Need… But Here's the Rest
- ⚡ Building GPT(2/3) from Scratch: Turning Theory into a Working Transformer
- 🧩 Understand, Then Build
- 🚀 Make ML Systems Ship Again
- 🏗️ The Production ML Survival Guide
- 🔍 Hard Lessons in Shipping AI/ML Projects
Check out more on my blog
I'm always looking to improve and tailor what I write and build. Whether you:
- Have a correction, question, or suggestion about something I wrote
- Want to discuss a paper, a derivation, or a from-scratch build
- Are interested in contributing to my open-source projects
- Are learning ML from first principles and want pointers
Feel free to reach out!
principles = {
"first_principles": "If I can't derive it, I don't understand it",
"build_from_scratch": "No black boxes - rebuild it layer by layer",
"write_to_learn": "Writing is how understanding forms, not how it's reported",
"timelessness": "Invest in what will still matter in 50 years",
"open_source": "Everything I build should be accessible",
"pragmatism": "Theory is great, but does it survive production?"
}