Blank Page Machine learning, from first principles.
A small, rigorous cohort for undergrads who want to derive the gradients, not just import the model. We work on a whiteboard, not a notebook.
Most ML courses leave you stuck between two traps.
Math that never meets a paper
Pure abstract theory, with no bridge to the models researchers actually build and publish.
Plug-and-play tutorials
You import a model and call .fit() without understanding its loss surface. It works until it doesn't.
Derive it, then build it
Start from a blank page: derive the gradients, check the conditioning of your covariance matrix, then write the algorithm from raw tensors.
Ten weeks, three phases.
Each phase pairs derivation-heavy lectures with a project on a real dataset.
Deterministic Learning & Linear Foundations
Weeks 1–3
- High-dimensional geometry
- SVD and PCA from scratch
- Ridge, Lasso, and the kernel trick
Probabilistic Machine Learning
Weeks 4–6
- MLE vs. MAP
- Gaussian processes, exponential families
- EM and latent variable models
Generative Modeling & Research Practice
Weeks 7–10
- Variational inference and the ELBO
- Paper-reproduction sprint
- Capstone, presented to the cohort
Built for undergrads heading toward research.
You'll get the most out of it if you…
- Are comfortable with linear algebra, vector calculus, and probability
- Want to understand why an algorithm works, not just that it does
- Are aiming for a research lab, a thesis, or graduate study
- Learn by discussing, and will show up with questions
It's probably not the right fit if you…
- Want a quick tour of ML libraries and APIs
- Prefer to watch recordings rather than join live
- Plan to have AI tools write your assignment code
From placement to capstone.
Take the placement check
30 questions in 30 minutes on the math you'll need. It isn't a rejection filter; it's how the cohort moves at research pace together.
Apply
Tell us where you are and what you're working toward. We reply within a few days.
Join your pod
You're placed in a Discord pod of 5–6 before week one, for doubt discussion between classes.
Learn, build, present
Two live classes a week, 10 peer-reviewed assignments, and a capstone presented to the cohort.
Taught at the board, not from the slides.
Led by a mentor pursuing an M.Sc. in Machine Learning at one of Europe's leading universities.
They bring academic and industry AI research experience from India, spanning computer vision, representation learning, and large-scale models. Office hours are hands-on: guiding derivations, unblocking stuck concepts, and holding you to the standard a real research lab will expect.
One fee, paid upfront. Money shouldn't be the reason you can't join.
Founding-cohort price. Regular price ₹5,000 + GST.
₹2,360 in total including ₹360 GST, paid upfront once you're accepted. Full refund if you cancel before the cohort starts.
- 20 live classes, taught at the board
- A Discord peer pod and office hours
- 10 assignments, reviewed by your pod and the instructor
- A capstone presented to the cohort
Need-based waiver
If the fee is a real barrier, it can be fully waived based on your financial background. Just tick the waiver box on the application.
Do the work, get money back
There are 10 assignments, and each one completed on time and on your own earns back ₹200. Finish all 10 and your ₹2,000 fee is refunded in full after the cohort ends. GST isn't refundable.
Common questions.
How much does it cost?
Cohort 1 is at a founding-cohort price of ₹2,000 plus 18% GST (₹2,360 in total), down from the regular ₹5,000, paid upfront once you're accepted. It can be fully waived based on financial background, and each of the 10 assignments you complete on time earns back ₹200, so finishing all 10 refunds the full ₹2,000 fee (GST isn't refundable). If you cancel before 1 November, you get a full refund. See fees and the Terms.
What's on the placement check?
30 multiple-choice questions in 30 minutes: 40% linear algebra, 30% vector calculus, 30% probability & statistics. No calculator. Each person gets a different set, and you verify your email with a one-time code before starting.
Can I retake the placement check?
Each email address gets one attempt. If you don't pass, you'll get a list of topics to review; reach out directly for a reset once you've worked through them.
What's the weekly commitment?
Two live classes a week, plus asynchronous project work and discussion in your peer pod. There are 10 assignments, each mapped to a phase topic.
Can I use AI tools?
Yes, to understand concepts faster. No, for writing your assignment code. The point is that you can build it yourself.
How are assignments graded?
Your peer pod reviews them first, then they're discussed in office hours.
What comes after Cohort 1?
Cohort 2 covers the current AI landscape: computer vision, transformers, vision-language and omni-modal models, and JEPA and world models. Cohort 1 is the prerequisite. See the Cohort 2 outline.
Twenty seats. Start with the placement check.
30 questions, 30 minutes. It tells you whether you're ready for week one.