You read my posts here. My real work lives on GitHub, and it's lonely over there.

Here's something that's been on my mind.

Every week, around 21 of you follow me here on dev.to. That genuinely means a lot, and I don't take it for granted. But here's the strange part: the posts you're reading are just the story. The actual thing, the code, the commits, the projects you can clone and run tonight, all of that lives on my GitHub. And GitHub is quiet.

People clone the repos. I can see it. But cloning and disappearing is like borrowing a book and never telling the author you liked it. So I'm going to do the thing I've been too shy to do: walk you through what I've actually built, and then straight-up ask you to follow me where it matters.

Let me show you what's over there.

ModelSentinel: the "can I trust this model?" toolkit

You train a model. Validation accuracy looks beautiful. You ship it. And then the quiet questions start: is it still accurate, or slowly degrading? Has the data drifted? When it says "92% confident," is it actually right 92% of the time?

ModelSentinel is my open-source Python library for everything that happens after model.predict(). Evaluation, data drift detection, calibration, data quality checks, all rolled into one Model Health Score you can alert on. I didn't demo it on toy data either. I turned it loose on my own deepfake detector and it scored the real model, real weights, real numbers. It's typed, tested, CI across Python 3.9 to 3.12, pip install and go.

DeepGuard: a deepfake detector that shows you where the fake is

Most deepfake tools hand you a score and ask you to trust it. DeepGuard draws a heatmap over the exact pixels that gave the fake away, usually the seams where a swapped face meets real skin. It's an EfficientNet-B4 model clearing 99% across two different fake-generation methods, with a full web app and a live Hugging Face demo with over 16k+ downloads.

And the rest of the shelf

There's more where that came from: medical imaging work, a RAG system for chatting with your PDFs, retrieval pipelines, and a few things still cooking. I ship new versions regularly. The pattern across all of it is the same: not notebooks, but real projects built like real software, with tests, docs, and honesty about their limits.

So here's the honest ask

I love that you read my writing. But the writing is the trailer. The movie is on GitHub.

If any of this made you curious, the single most useful thing you can do is follow me on GitHub. Not for a vanity number. It's because that's where I actually build in the open, where the new versions land first, where you can open an issue, star a repo, fork it, or just watch how a project evolves commit by commit. Following here tells me a post landed. Following there tells me the work landed, and honestly, that's the signal that keeps me building.

So if you've ever cloned one of my repos and quietly closed the tab, this is me waving.

👉 Follow me on GitHub. Star the repo that's useful to you. Open an issue if something breaks. I answer every one.

That's the whole ask. Thanks for reading, and I'll see you in the commit log.