Maths and Python for AI
Take this if you want to build AI from first principles and would rather fix the foundations now than hit a wall in month six.
What you’ll be able to do
You can differentiate and optimise a function by hand and in code, manipulate vectors and matrices fluently with NumPy, reason correctly about probability and uncertainty, and write clean, tested Python with version control.
What it assumes
School algebra and the ability to read a graph. No calculus and no programming experience assumed — both are built from the start.
- Price
- ₹1,499
- Length
- 20 weeks
- Sessions
- 20
- Effort
- ~7 h/week · 140 h total
- For
- Class 9–10 · Class 11–12 · College · Working professional
Adds it to My courses on your dashboard so you can find it again. Payment is not open yet — nothing is charged.
What’s taught, in order
- 01
Functions, calculus and optimisation
Derivatives, the chain rule, integration and multivariable calculus, each introduced by the machine learning problem that needs it. Optimisation arrives as the destination the whole strand was walking toward.
- 02
Linear algebra as data and transformation
Vectors, matrices, spaces, eigenvectors and the singular value decomposition, taught geometrically before algebraically. This is the strand that makes every later library call legible.
- 03
Probability, statistics and information
Distributions, Bayes, estimation, hypothesis testing and entropy. Enough to reason honestly about uncertainty, which is what separates a careful practitioner from a confident one.
- 04
Python, from syntax to engineering
Data structures, functions, objects, environments, testing and Git, then NumPy, Pandas, Matplotlib and SQL. Software hygiene is taught alongside, not left as an afterthought.
- 05
Consolidation and a capstone
Two weeks that spend nothing new: a programming capstone that uses the entire toolkit at once, and a self-assessment that tells you honestly whether you are ready for the next phase.
What you build
Build a complete analysis of a dataset you choose, from raw file to written conclusion: load and clean it with Pandas, explore it with real statistics rather than eyeballing, implement one optimisation routine by hand rather than calling a library, visualise the result, and ship it as a tested, version-controlled repository with a README somebody else could follow.
Check it yourself against this
- The repository runs from a clean checkout following only your README.
- At least one optimisation or numerical routine is implemented by you, not imported.
- Your statistical claims include a measure of uncertainty, not just point estimates.
- Tests exist for the non-trivial functions and they pass.
- The commit history shows incremental work rather than one upload.
- The written conclusion states what the data cannot tell you.
Then answer this
Which single idea from these twenty weeks turned out to be load-bearing for the most other ideas, and why that one?
Two or three sentences, in your own words. If you cannot, the course is not finished — go back to the module it came from.
What this course does not do
This phase deliberately contains no machine learning models. You finish it able to build the machinery that models are made of, having trained nothing — which feels slow, and is the reason the phase exists. If you want a model working this month, start with a sprint instead.
Where this leads
The Foundations phase page sets out the full twenty-week week-by-week breakdown, the prerequisites in detail, and what the phase deliberately leaves out, so you can check the plan before committing to five months.
FAQ
- Twenty weeks with no machine learning at all — really?
- Really, and it is the most common objection. The alternative is meeting gradients, eigenvectors and probability for the first time while also learning a model, which is where most self-taught paths stall. The free courses and the sprints exist precisely so you can have an early win alongside this.
- I already know Python. Can I skip that strand?
- You can move through it quickly, and the self-assessment in the final week is designed to tell you where you genuinely are. Most people who describe themselves as knowing Python have not met environments, testing or Git properly, and those three are what the later phases assume.
- How does this compare to a university first-year maths course?
- Narrower and more applied. There are no proofs and no general theory; every topic is chosen because machine learning needs it and dropped where it stops being needed. That makes it a poor substitute for a formal syllabus and a good complement to one.
Other courses
This is a standalone course. It does not add weeks to the 78-week curriculum or change your roadmap. If you want the whole path instead, the full programme is ₹399/month · ₹2,000/year — see what it covers.