The ML Debugging Training
Online 5-week bootcamp to teach you how to start debugging ML systems like a pro.
1st cohort, only 5 spots available! 🚀
ML debugging is hard ⛰️
Ever trained a model to convergence, only to see a metric scream that something’s wrong? Spent days chasing a bug, to discover it was a missed batchnorm.eval()? Or randomly started getting NaNs in your overnight training 2 hours in?Debugging is always tricky, but ML debugging is a special kind of pain: complex code, stochastic behavior, messy data, limited observability, you name it. You know something’s broken in the pipeline, but where do you even start?Everyone in ML faces this, yet practical guidance is scarce. It’s like this silent, universal problem that no one seems to be trying to fix.
Not anymore.
Introducing the ML Debugging Training. A 5-week online bootcamp that will teach you the skills and the framework to debug machine learning systems efficiently, tailored to the needs of your own projects.No more stumbling around until you find the bug. No more feeling lost about where to start looking. No more burning days chasing dead ends. Gain the skills, confidence, and clear framework to debug any ML system 💪.The training distills years of hard-earned lessons from my PhD and professional ML career, and is structured into 5 weeks, covering the following:
1. Preparation
No more falling head-first without preparation 💪. Learn which steps you take before you even get a bug, to put you in the best position possible when the debugging starts.
2. Using a Debugger
From zero to hero 🛠️ . Use a debugger to stop execution and inspect variables, step through lines, move up and down the stack, add conditional breakpoints, and everything else needed to be an effective bug hunter.
3. Hard Failures
Efficiently troubleshoot bugs with a clear point of failure. From symptom to root cause with the least effort possible, with several examples for you to put what you learn into practice.
4. Soft Failures
Have a clear framework to follow even when the symptoms don't point to a clear starting point (think weird loss curves, or poor test performance). Learn how to experiment effectively to save you hours of debugging.
5. Reflection
Use every debugging experience as a learning opportunity to improve your craft. Start a positive feedback loop between debugging sessions and all the previous steps.
Live sessions every Wednesday 16:00-18:00 CEST, starting September 24 🗓️.
This course is for you if...
You already have some ML experience
You have worked on a few different ML projects, and are familiar with the basic theory.
You're willing to put in the work
You have the availability to spend at least 2-3 hours per week on the training outside the live sessions. This is not a skill you can learn by just hearing me talk 😊.
You want to ship ML projects faster
From zero to MVP to production as fast as possible. Spending time in the right things, to optimize time-to-deliver.
You want ML debugging to hurt less
You want to feel confident and prepared when a bug appears in your ML pipeline. No more stumbling around, feeling lost.
This is not a beginner course. We will not go through any machine learning basics 😉.
Confident Debugging Guaranteed
I will do my absolute best to help every single student that enrolls to up their debugging game. That said, if you attend all sessions, do the work, and at the end of the course still feel like your investment wasn't worth it, I'll refund you in full.
I know how it feels to be stuck.
During my PhD I did most things on my own, from scratch. RL loops, on-the-fly synthetic data, async pipelines. Each training run took hours (sometimes days). And I wasn't really good at it 🙈. I didn't really have the right tools to deal with all that complexity.This translated to many bugs. Silent bugs, nasty bugs, silly bugs, all of them. I remember spending an entire week debugging why my model wasn't learning, only to later realize that every parallel data loader was serving the same batch because of a silly mistake 🤦.Failing solo like that hurt, but it also put me in a really good position to improve. Whenever I got a bug, I knew I was the one that caused it. And I owned the codebase, so it was up to me to not do that again.That, together with a lot of working experience alongside senior engineers when I joined industry had an immense impact on my skills. After some time I could see how my bug count (and debug time) really went down. Not only that, but also I started having this mindset that no bug is too hard, and that every bug is a new learning opportunity to improve your craft.This training distills that hard road so you can skip the multi-year slog. It's exactly what I wish I had when I was struggling with ML debugging and development.
You won't be doing this alone!
Community & Support
You will get access to a private group where you can ask questions, share debugging wins, and get help when you’re stuck.Miss a session? No worries, all of them are recorded and available for 30 days.
Join the ML Debugging Training 🚀
• Weekly live sessions every Wednesday, 16:00–18:00 CEST, on mastering ML debugging.• Hands-on exercises to practice and adapt the framework to your work.• Private community access to share wins, ask questions, and get feedback.• Risk-free guarantee: full refund if you complete the program and find it wasn’t worth it.
SEK 2.490,00
5 spots only! First cohort starting September 24.
FAQ
How is the course structured?
We will have weekly sessions where you will be expected to participate (think camera on, keyboard ready 😄). They will consist of me teaching, along with coding and debugging exercises for you to do. There will also be some take-home tasks to help tailor what you learn to your specific projects.
*
What if I can't attend a session?
All sessions are recorded and will be available for 30 days.
*
What if I run into bugs you don't cover in the course?
That’s the beauty! You won’t just memorize fixes, you’ll learn the processes to debug anything, and how to become better and better over time in the contexts that you work in.
*
Is this relevant if I only use pre-trained models/LLMs/etc?
Absolutely. Pre-trained models still break (and the pipelines you build around them), so you need to know how to spot and fix their issues.
*
Other questions?
Send me a DM in LinkedIn or an email and I'll answer ASAP 👍
© Juliano Labs. All rights reserved.