Learn Machine Learning with JavaScript
You already think in arrays, state, and async. Machine learning uses the same patterns under different names. Tensorcraft maps them with 55 JavaScript-to-ML bridges and teaches them through one story: 11 modules aboard a station in deep orbit, written in TypeScript, run in your browser.
Module 1 of Deep Orbit, the live course, is free: no account, no credit card.
How it works
A bridge
You already know more ML than you think: every concept in the course is anchored to JavaScript you write today, across 55 bridges, each labeled where the analogy breaks. Here is one.
Component state Model parameters (weights)
Bridge 2 of 55const [count, setCount] = useState(0);model.getWeights().forEach(w => console.log(w.shape))bridge: Component state and model weights are the internal values the system owns and mutates in response to inputs. State updates from user events; weights update from training data.
// where it breaks
State updates instantly on user events with deterministic transitions; weights update slowly via gradient descent over many iterations and the update rule is stochastic. State is small and hand-written; weights are millions of floats with no semantic meaning per index.
Run it
A real 4 KB neural network scores six archived sensor readings in your tab, the same demo that ends Module 1. Watch sample 7741: 51 percent is not an answer.
loading model (4 KB) into your browser…
The road
The whole course, module by module. Module 1 is free in full: the lessons, the exercises, and the demo above. No account, no credit card.
11 modules · 56 lessons · one continuous story · ends with a capstone model running live in the browser
Module 1: Foundations: Tensors & Data
FreeTensors, shapes, and the operations that reshape them, starting from the Float32Array you already use.
Start here, free, no account →5 lessons · ~107 minModule 2: Neural Networks: Layers & Architecture
Build a classifier one piece at a time: a single neuron, the activation that bends it, and layers stacked into an architecture.
5 lessons · ~129 minModule 3: Training: Loss, Gradients & Optimization
Loss functions, gradient descent, the training loop that ties them together, and the evaluation that decides whether a model is good enough to ship.
5 lessons · ~157 minModule 4: Browser ML Runtimes
Run a trained model in the browser: TensorFlow.js setup, model loading and caching, real-time inference, and how TensorFlow.js, ONNX Runtime Web, and Transformers.js compare.
6 lessons · ~162 minModule 5: Signal Processing
Turn raw time-series data into features the Fourier Transform, spectrograms, and windowing can expose.
5 lessons · ~131 minModule 6: Recurrent Networks
Recurrent networks that carry state across time: RNNs, LSTMs, and GRUs, mapped onto the reducers and cache expiry you already write.
5 lessons · ~122 minModule 7: Anomaly Detection
Autoencoders that learn what normal looks like, then flag whatever they cannot rebuild, using reconstruction error and a tuned threshold.
5 lessons · ~174 minModule 8: Advanced Time-Series
Attention, multi-step forecasting, Kalman filters, and confidence-weighted ensembles, for data that does not repeat on a fixed schedule.
5 lessons · ~135 minModule 9: LLM Integration
Add language models to the pipeline: prompt engineering, retrieval over your own logs, and orchestrating more than one model.
5 lessons · ~195 minModule 10: Optimization & Edge
Profiling, quantization, pruning, caching, and WebGPU acceleration, treated as a measured tradeoff between model size and accuracy.
5 lessons · ~200 minModule 11: Capstone: Deploy DEEP SCAN
Assemble the finished system: architecture, a deployment pipeline, monitoring, and the documentation that ships with it.
5 lessons · ~232 min
Deep Orbit is written as one continuous story: you join the crew of a research station and learn each technique when the plot needs it. The people you work with:
- Commander MirzaMentor
- Dr. FarahAstrophysicist
- Chief NazariSkeptic
- Ensign DemirCommunications
- ARIAShip AI
No Python required
The whole course runs TensorFlow.js. You write TypeScript, the browser runs it, and the model you finish with works anywhere a browser does. The same code runs under Node when you want it on a server. There are no notebooks to set up, nothing to pip install, and no GPU drivers to fight.
Coming soon
4 more courses are planned. No release dates announced yet:
Common Questions
Can I learn machine learning with JavaScript?
Yes. Tensorcraft teaches machine learning entirely in JavaScript using the TensorFlow.js library. You will build models, then train and run them in the browser right in the lessons; the same code runs under Node. No Python required. The live world, Deep Orbit, covers topics like neural networks, recurrent networks (LSTMs), signal processing, and attention. More topics are covered in the upcoming worlds.
What if I need Python or PyTorch?
Then skip Tensorcraft. It is entirely taught with TensorFlow.js and intended as a fun and easy way for JavaScript engineers to break into machine learning.
How much math do I need?
None upfront. You just need JavaScript knowledge and comfort with concepts like functions, arrays, and async. The math will be introduced when you need it, in optional expandable drawers.
Is there a video track?
No. Lessons are scenes written in text, with code snippets alongside. Completing exercises opens the gate to new lessons. Short video scenes are on the roadmap: subscribe to updates on the Worlds page to hear when they land.
How long does Deep Orbit take?
Deep Orbit consists of 56 lessons across 11 modules. Since it's story-based, it's meant to be tackled in order. Each module's time estimates are listed on the pricing page and the Deep Orbit course page.
Is Tensorcraft free?
Module 1 of Deep Orbit is free, no account or credit card is required. However, the full course costs $59. You will make a one-time payment and get lifetime access to your purchased world.
What if it turns out not to be for me?
Module 1 is free. The full course is paid: after purchase there's a 5-day refund guarantee, no questions asked.
Module 1 is free. After that, each course is a one-time purchase: see pricing for what each plan includes.