Understand LLMs in
one weekend, not one semester.
68 hands-on exercises where you send real prompts to real AI models. The concepts that take 20 hours to learn from videos take 3 hours when you do them yourself.
You'll need a free Gemini API key from Google AI Studio. Takes 2 minutes.
Hello, Intelligence
At its core, an LLM is an autocomplete machine on steroids. It predicts the most likely next word based on patterns.
Type "Jana Gana M" and let the model finish the word.
Not a video course. A playground.
You learn by doing things, not watching someone else do them. Every concept has an exercise you run yourself.
Real models, real prompts
Your exercises use the same Gemini API that powers Google products. Live calls, not simulations or pre-recorded outputs.
Tweak every parameter
Temperature, Top-K, Top-P, system instructions, stop sequences, thinking budget. Change them live and see the difference.
We show what fails
We told AI we solved P=NP. It believed us. Hallucinations, sycophancy, and bias aren't footnotes here — they're entire lessons.
Instant validation
Every exercise has built-in checks. You know in seconds if your prompt worked. No waiting for a grader or a certificate queue.
Learn by breaking things
Set temperature to 2.0 and watch gibberish. Ask it to count letters in 'strawberry' and watch it fail. That's the lesson.
Your key, your quota
Use your free Google AI Studio key. We never see your prompts or your API key. Your conversations stay between you and Google.
Where you are → Where you'll be
After 68 exercises, this is the difference.
I use ChatGPT every day
I understand how LLMs actually work under the hood
I can't explain temperature or tokens in an interview
I can set up the right config for any LLM task and explain why
I copy-paste prompts from the internet
I craft prompts using zero-shot, few-shot, and chain-of-thought techniques
I trust whatever the AI tells me
I know when models hallucinate, why they agree with nonsense, and how to spot bias
Students get it in hours, not weeks.
From "I use ChatGPT" to "I understand how LLMs work" — in one weekend.
"I've watched 10 hours of AI tutorials on YouTube. I learned more in the first 30 minutes of TinkerLLM than all of them combined."
"The hallucinations lesson blew my mind. I told the AI I solved P=NP and it congratulated me. Now I understand why you can't trust model outputs blindly."
"I finally understand what temperature, tokens, and system instructions actually do. Not because someone explained it — because I changed them and saw the difference."
Testimonials are placeholder — real student feedback incoming.
18 lessons. Zero fluff.
From your first prompt to hallucination detection. Here are 5 you'll love.
Hello, Intelligence
FreeYour first conversation with an LLM. Autocomplete, knowledge retrieval, and the hallucination you didn't expect.
System Instructions
The constitution of your model. Build personas, enforce JSON output, create guardrails that block off-topic requests.
Few-Shot Prompting
The single most powerful prompting technique. Teach the model any format using just 2-3 examples.
Hallucinations & Sycophancy
We told AI we solved the hardest problem in computer science. It congratulated us. Learn why models lie and agree.
Safety, Ethics & Alignment
Jailbreaking, implicit bias, and the values hardcoded into every model. Test the guardrails yourself.
One price. No surprises.
No subscription. No recurring fees. No upsells. Pay once, tinker forever.
No credit card required
- Lessons 1-2 (26 exercises)
- Full playground access
- Autocomplete, tokens, knowledge retrieval
- Bring your own Gemini API key
- Temperature, prompting, system instructions
- Hallucinations, safety, vision, coding
- Completion certificate
One-time. No subscription.
- All 18 lessons (68 exercises)
- Temperature, Top-K, Top-P controls
- System instructions, prompting strategies
- Chain of thought, few-shot, zero-shot
- Hallucinations, sycophancy, AI safety
- Vision, coding, JSON mode
- Completion certificate
International: $9 USD
You bring your own Gemini API key (free from Google AI Studio). We don't charge per message or per API call. Your cost is ₹499. That's it.
Ready to tinker?
26 exercises free. No credit card. No catch. Just open the playground and start.
Open the Playground — FreeStop watching. Start tinkering.