We got tired of watching.

Every AI course we found was the same. Someone talks at a camera for 20 hours. They show their screen. You watch. You maybe take notes. Then you close the tab and forget 90% of it.

We wanted something different. A place where you actually send prompts to real models. Where you change the temperature and see what happens. Where you break things on purpose and learn why they broke.

So we built TinkerLLM. A playground with a curriculum. 68 exercises that teach you LLM fundamentals by making you do them, not watch someone else do them.

Built by Kalvium Labs

TinkerLLM is built by Kalvium Labs, an AI product studio with 200+ engineers. We build AI products for startups across the US, Gulf, and India. Every project is supervised by our CTO, Anil Gulecha (ex-HackerRank, ex-Google).

TinkerLLM started as an internal training tool. We needed our own engineers to understand LLM parameters, prompting strategies, and failure modes before they shipped AI products for clients. The exercises worked so well that we turned it into a product.

The curriculum comes from real production experience. When we teach you about hallucinations, it's because we've debugged them in client projects. When we teach you about system instructions, it's because we use them every day to build AI products that actually work.

What we believe

• You learn by doing, not watching.
• Knowing what AI gets wrong is as important as knowing what it gets right.
• Every concept should be testable in a playground, or it's not worth teaching.
• AI education should cost less than a meal, not more than a semester.