Project course · 14 lessons · 4 modules

Build an LLM System in Your Browser

Build the model, runtime, serving layer, and React application that come together as one working local chatbot.

Restoring your place…

Course guide

Build the whole path, or jump to the part you need

Start with Character RNNs Open coding workspace

Fit, prerequisites, outcome, and runtimeCourse details

Best for: Python developers who know NumPy, loss, and gradients and want to see the whole LLM path.
Before you start: Know basic linear algebra and ML, or use the two foundation courses as refreshers.
Outcome: You'll end up with a tested chatbot that saves data locally, runs a real model, loads the RNN checkpoint you trained, and uses browser adapters checked against the same behavior as your Python code.
Runtime: Python + NumPy + ReactRuns locally in your browser. Lesson files accumulate in the saved Browser Chat project.Python lessons are checked separately; the capstone runs tested browser adapters and an optional local model download. No API key is required.

Already comfortable with the model basics?

You can open any lesson directly and return to earlier material when you need a refresher. The complete Browser Chat build still requires the full project.

Module 016 lessons

Model Foundations

Build the full number-crunching path from tokenization and learned representations to causal language modeling and local inference.

Outcome

You'll have a character model you trained and a test setup for evaluating a frozen local model.

Module 022 lessons

Inference Runtime

Build prefill, decode, KV-cache accounting, admission, and continuous batching for autoregressive inference.

Outcome

You'll have a runtime model you can run, with clear timing for each phase, memory use, and scheduling behavior.

Module 032 lessons

LLM Serving

Build clear rules for streaming, cancellation, retries, failures, and observability around model generation.

Outcome

You'll have a repeatable serving layer and a failure-injection test setup for the browser chatbot.

Module 044 lessons

Chat Integration

Connect generation events to a React state machine that can recover from problems, manage context, save data, and stay accessible.

Outcome

You'll have a polished streaming chat interface with recovery tools, context controls, and a clear view of generation state.

Browser Chat

Combine the model, runtime, streaming transport, and React client into one local chatbot.