80topics
Game Theory Chat
Strategy, equilibrium and incentives, from the prisoner’s dilemma to mechanism design.
Free · no account · open source
Ask the hard question. Get the worked answer.
Game theory, data science, reinforcement learning, linear algebra, the web platform and the Islamic Golden Age. Each assistant retrieves curated explanations with the math shown, and answers in your browser.
Your questions are processed on your device. The code is MIT-licensed.
max(0, x), is the function this site is named after: silent while the input is negative, then straight up. Push x right and each subject switches on.Pick a subject
Six assistants, each with its own curated knowledge base
50topics
Golden Age Inquiry
Scholars, institutions and ideas of the Islamic Golden Age: algebra, optics, medicine and philosophy.
Try askingAl-KhwarizmiHouse of WisdomIbn al-Haytham (Alhazen)
41topics
Data Science Chat
Statistics, evaluation and machine-learning concepts with the math shown.
Try askingTrain/Test SplitLogistic RegressionProbability Distributions
16topics
Reinforcement Learning Chat
Rewards, value functions, bandits, Q-learning and policy gradients.
16topics
Linear Algebra Chat
Vectors, matrices, eigenvalues and projections: the algebra behind machine learning.
Try askingEigenvalues and eigenvectorsDot productLeast squares
16topics
Web Platform Chat
Browser JavaScript, HTTP caching, WebAssembly, workers and local storage.
Try askingService workerEvent loopWebAssembly
Assistants retrieve and assemble curated fragments. They can misunderstand a question, do not execute code and cannot solve arbitrary exercises.
How it works
From your question to an answer, one step at a time
- 1
Retrieve
Keyword matching and field-weighted BM25 find candidate topics instantly. An optional ~22 MB sentence model adds broader wording matches.
- 2
Read the signals
Entities, intent cues and ranked scores merge into one structured packet, all on your device.
- 3
Decide
A 13,079-parameter policy network chooses how to respond: answer, clarify, compare or follow up.
- 4
Compose
Curated fragments are joined with connectors into a precise answer. Nothing is generated.
- 5
Render
KaTeX typesets the math and session memory keeps follow-ups in context.
Interactive tools
Change a number, watch the algorithm respond
- Neural Network ExplorerAdjust weights and activations and watch a forward pass move through every neuron.→
- Activation Functions ExplorerPlot ReLU, Sigmoid, Tanh, GELU and more with their derivatives, and see why slope decides whether learning continues.→
- Gradient Descent LabClick a starting point, set the learning rate and follow each step down a loss surface.→
- Backpropagation VisualizerWalk the chain rule backwards through a 2→2→1 network with every gradient derived.→
- K-Means Clustering PlaygroundAdd points, choose k and step through assignment and centroid updates as inertia shrinks.→
- Decision Tree ExplorerGrow a tree one split at a time and watch Gini impurity fall, then overfitting appear.→
Worked examples
Latest guides
- Dot Product and Cosine Similarity, Step by StepWork through a dot product, vector norms, and cosine similarity, including the zero-vector case that retrieval code must handle.
- Data Leakage: Split by Entity Before You TrainWhy random row splits overstate performance on repeated users, devices, or experiments, and how grouped validation…
- Least Squares Without Computing an InverseDerive a two-parameter fit, check its residuals, and understand why practical solvers use QR or SVD instead of an…
From the makers
Go further than the site
Book · 224 pages
How to Remain Valuable When Intelligence Becomes Cheap
A 224-page practical guide to the human, economic and strategic advantages that stay valuable as AI takes on more of the world's cognitive work.
Toolkit · $29
On-Device Chatbot Builder Toolkit
You have seen what local, privacy-first AI can do here. The toolkit hands you the architecture behind it: frameworks, templates and evaluation guides covering the full pipeline, from raw input to validated response.
Get the toolkit
Ebook · $1
Fringe Learning: Resource-Efficient RL for Edge ML
Build your own custom AI systems with practical reinforcement learning methods designed for resource-constrained edge machine learning.
Get the ebookAbout
An open-source experiment in simpler chatbots
ReLU.chat is built by Yunus Emre Vurgun to find simpler, faster ways to build chatbots, without creating an LLM or leaning on pre-LLM chatbot logic.
The goal is to combine sentence transformers, BM25 sparse retrieval, a reinforcement-learned policy network and fragment-based composition into a system that runs in the browser. No chat API or API key is needed, so your questions stay on your device. Hosting, optional model downloads and external links still use the network.