Learn AI the way it's actually built, not just described.
Valmiki AI Lab is a hands-on studio where students move from first principles to working systems — writing the math, training the models, and shipping real projects alongside practitioners.
Named for a teacher, built for practice.
Valmiki AI Lab started from a simple frustration: most AI courses teach students to use tools, not to understand or build them. Our sessions are run like a workshop — small groups, a whiteboard, and a dataset, working through problems until the model in front of you actually makes sense.
Every program is anchored to a project you keep: a classifier you trained, an agent you shipped, a paper you can explain line by line. Students leave with a portfolio, not just a certificate.
- ✓ Small cohorts, taught live by working ML engineers and researchers
- ✓ Every course ends in a project you build and present, not a quiz
- ✓ Lab hours outside class for one-on-one debugging and code review
Six paths, one lab.
Start where you are. Each program is cohort-based, runs on a fixed weekly schedule, and ends with a project review.
Python & Math for AI
Linear algebra, probability, and Python built around the operations you'll actually use to train models — not abstract theory.
Machine Learning Fundamentals
Regression to gradient boosting: build, evaluate, and debug models on real datasets, from raw CSV to a working pipeline.
Deep Learning & Neural Networks
Backpropagation by hand, then in PyTorch. Train image and text models and learn why they fail before you learn how to fix them.
Large Language Models & Transformers
Attention, tokenization, fine-tuning and evaluation. Build a small language model from scratch and fine-tune an open model on your own data.
Building AI Agents & Products
Ship a real application: retrieval, tool use, evaluation, and the unglamorous work of making an AI feature reliable in production.
Capstone Research Lab
Pick an open problem, work with a mentor, and produce a project write-up worth putting in a portfolio or submitting to a workshop.
Built like a lab, not a lecture hall.
The format is the same across every level: small groups, live coding, and a project that forces the concept to become real.
Build before you're told how
Sessions open with a problem, not a slide deck. You try, get stuck, and the concept lands when the explanation actually answers your question.
Every model, from scratch once
Before you use a library, you implement the core idea in plain code, so the library stops being a black box.
Weekly one-on-one review
A mentor reads your code and your results every week — not just at the end of the cohort.
Real, messy datasets
No toy CSVs. You clean, question, and argue with data the way you will on the job.
Projects, not certificates
What you leave with is a repository and a demo, something you can show, not just claim.
Small cohorts, always
We cap every cohort so mentors know your work by name, not by student ID.
Practitioners, not presenters.
Mentors are working engineers and researchers who teach part-time because they care about how the field gets taught.
Ananya Rao
Lead Mentor, Deep Learning
Ten years building computer vision systems in production; leads the Deep Learning and LLM tracks.
Rohan Mehta
Mentor, ML Fundamentals
Former data scientist turned educator, focused on making the math behind ML click for beginners.
Sana Iqbal
Mentor, Applied AI & Agents
Builds AI products for a living; runs the Capstone Research Lab and the agents studio track.
What students say after they ship.
"I'd used ML libraries for two years and never once understood backprop until I had to write it myself here."
"The weekly one-on-ones caught mistakes in my code that would have taken me weeks to find alone."
Start your application.
Tell us where you are and where you want to get to. We'll follow up with the right program and the next cohort's start date.