Who it’s for
Learners looking for a hands-on introduction to how language models work, from the first concepts to a working project.
LIVE, WITH AN INSTRUCTOR
Build Your First Language Model
Fine-tune a real model. Ship a demo in 6 weeks.
For beginners with no ML background. Use guided notebooks to fine-tune Needle, a 26-million-parameter model, and build a web demo you can share.
12 live sessions6 weeks
FIND YOUR FIT
Is this program for you?
Before you start
No prior machine learning background is required. The course introduces the concepts and uses guided notebooks.
MAKE IT REAL
Your model. Your task. A working demo.
Choose a focused task, prepare examples, and fine-tune Needle, a 26-million-parameter open-source model. Wrap it in a simple web demo and share it on demo day.
THE TAKEAWAY
What you’ll be able to do.
Explain how text, tokens and attention fit together in a language model.
Explore a real model and its data in Google Colab.
Create a focused dataset and fine-tune Needle for your chosen task.
Build a clickable demo and present what you made.
MEET YOUR INSTRUCTOR
HOW THE PROGRAM WORKS
Your weekly learning plan.
Explore each week to see the sessions, topics, and teaching dates. All times are in IST.
WEEK 01Meet your model and your AI labExplore next-token prediction and run Needle in Google Colab.
How Machines Learn to "Understand" Text
The big idea behind every language model: predicting the next word, over and over, at huge scale. What Needle, a real 26-million-parameter model, tells us about how small "useful" can actually be.
Your Free AI Lab
Setting up Google Colab, running your first notebook, and loading Needle to watch it pick a tool and fill in the arguments.
WEEK 02Turn language into training dataUnderstand tokenization and inspect real tool-calling examples.
Why Computers Can't Read (And What We Do About It)
Tokenization explained simply: breaking text into pieces a model can learn from, and why vocabulary size matters.
Exploring a Real Model's Data
What training data actually looks like, using Needle's own tool-calling examples as a real, inspectable dataset.
WEEK 03See attention in actionExplore context and build a tiny attention block.
Attention, Without the Math
The one idea behind every modern model: letting each word "look at" the others for context. Needle makes an unusually good teaching example, since its architecture drops the usual extra layers and keeps only attention. There's less to explain.
Building a Tiny Attention Block
Assembling a small attention block in code, guided line by line. You don't need to derive it, just see it work and inspect it.
WEEK 04Understand how a model learnsConnect loss, learning rates and epochs to a training run.
What "Training" Really Means
Loss, learning rate, and epochs, explained as "how a model finds out it's wrong and gets a little better every time." Plain language, with a training run you can actually watch.
How a Real 26M-Parameter Model Gets Made
Needle was pretrained on 200 billion tokens over about 27 hours, then specialized in roughly 45 minutes of fine-tuning. Why the second number is the one that matters for this course, and for you.
WEEK 05Make the model your ownDesign a task, prepare a dataset and fine-tune the model.
Designing Your Own Task
Choosing something specific and useful for your model to do, and building a small, focused dataset of examples for it.
Fine-Tuning, For Real
Running Needle's own fine-tuning playground on your data, and watching your general-purpose model become specific to your task.
WEEK 06Build your demo. Share your work.Create a clickable web demo and present it on demo day.
Making It Runnable
Wrapping your fine-tuned model in a simple, clickable web demo anyone can try. The same instinct behind Needle's own phone-ready design.
Demo Day
Present what your model does, how you built it, and what you'd try next, to classmates, faculty, or anyone willing to test it live.
BEFORE YOU JOIN
A few practical questions.
Can I revisit the sessions?
Session recordings are included with the program.
Where do I find my program after registering?
Sign in and open your dashboard to access your enrolled programs.
Where can I check the refund terms?
Read the refund policy before enrolling. For questions about your situation, contact hello@gradlabs.in.
YOUR NEXT STEP
Ready to join this cohort?
Review the current fee and reserve your place, or ask us a question before you decide.