AI Learning Lab

While learning AI, I often found myself understanding the theory but not really having an intuition for what was happening inside a model. I wanted a way to learn by actually interacting with the models instead of just reading about them.

That idea became AI Learning Lab.

What it does

AI Learning Lab is an interactive AI learning environment that runs models directly in the browser.

You can experiment with embeddings, gradients, RNNs, attention, and transformers and see the actual computations happen. The browser-based exercises don't need an API key or a server to run.

How we built it

The project is built with Next.js and TypeScript. I use Transformers.js and ONNX Runtime Web for browser inference, with WebGPU and WASM support.

The learning exercises have separate computation engines from the UI, which makes them easier to test. Models and user progress are also cached locally so the browser experience can work offline.

Challenges we ran into

The biggest challenge was making the experiments actually use real model computations instead of just showing pre-made animations.

Running models in a browser also meant dealing with model size, memory, WebGPU support, WASM fallbacks, and caching across different devices.

Accomplishments that we're proud of

I'm most proud that the interactive parts are backed by real computation. When the lab shows an embedding, gradient, attention weight, or hidden state, it comes from the model actually running.

I also got the browser experience working offline after the initial model download.

What I learned

Building AI for the browser made me think much more about model size, memory, inference speed, and hardware limitations.

It also reinforced how useful it is to keep the model logic separate from the UI. It made the project much easier to test and extend.

What's next for AI Learning Lab

I want to add more experiments around how modern AI systems actually work, especially topics that are difficult to understand from theory alone.

I'd also like to explore more ways to run useful AI locally, including model optimization and quantization.

Built With

• cloud
• docker
• google
• indexeddb
• mongodb
• next.js
• onnx-runtime-web
• react
• tailwind-css
• transformers.js
• typescript
• vitest
• wasm
• webgpu
• webllm
• zustand