# Ornix > Ornix is a learning planner for self-study: it plans a dated, week-by-week roadmap around a learner's hours, shows the next goal each day, and keeps a learning memory that Claude, ChatGPT and other MCP clients can read. Connect over MCP at https://mcp.ornix.io/mcp; setup: https://ornix.io/docs/mcp.md. The home page, pricing, the roadmap gallery, docs, and public roadmaps and notes are available as Markdown: append .md or send Accept: text/markdown. ## Product - [Ornix overview](https://ornix.io/index.md): What Ornix does, how a learner starts and how to connect an AI client. - [Pricing](https://ornix.io/pricing.md): The tiers, what each one includes and the pricing questions. - [Ready roadmaps](https://ornix.io/start.md): Ready roadmaps, picked by hand, that a learner can follow as their own copy. ## Featured roadmaps - [Large Language Models, End to End](https://ornix.io/u/omar/large-language-models-end-to-end.md): 12 weeks, by @omar. For developers who know Python and PyTorch: 12 weeks to understand how LLMs work and are trained. You build a tokenizer, a GPT, a LoRA fine-tune, an eval harness and a tiny pretrain, tune and evaluate pipeline. - [Machine Learning from First Principles](https://ornix.io/u/omar/machine-learning-first-principles.md): 12 weeks, by @omar. Twelve weeks for a programmer with high-school math. Understand classical ML by writing regression, trees, boosting, PCA and k-means in NumPy, then ship an honest end-to-end tabular project with a model card. - [Deep Learning from Scratch: backprop to diffusion](https://ornix.io/u/omar/deep-learning-from-scratch.md): 12 weeks, by @omar. For programmers who know Python and a little ML: 12 weeks, 7 to 10 hours a week, from hand-written autograd to a diffusion model you train yourself. Four projects, no black boxes. - [Building AI Agents: tools, memory and evals](https://ornix.io/u/omar/building-ai-agents.md): 10 weeks, by @omar. For software engineers who already call an LLM API. In 10 weeks, hand-roll the agent loop, add retrieval, memory, an MCP server, evals and safety, and ship a deployed agent with tracing, cost caps and a kill switch. ## Notes Explainers published with the featured roadmaps. - [BPE by hand](https://ornix.io/u/omar/large-language-models-end-to-end/bpe-by-hand.md): A language model never sees text. It sees integers. The tokenizer is the part that decides which integers, and a surprising number of "model" failures (bad… - [Attention, one matrix at a time](https://ornix.io/u/omar/large-language-models-end-to-end/attention-one-matrix-at-a-time.md): Attention is a soft dictionary lookup, and most of what makes it feel hard is bookkeeping about shapes. The cure is to do it once with numbers small enough to… - [Reading a loss curve](https://ornix.io/u/omar/large-language-models-end-to-end/reading-a-loss-curve.md): A loss curve is the cheapest instrument you own. Read it well and it tells you whether the model is learning, whether the code is wrong, and which knob to… - [Why scaling laws bend (Chinchilla)](https://ornix.io/u/omar/large-language-models-end-to-end/why-scaling-laws-bend.md): Scaling laws are the most useful empirical result in the field, and the most often misquoted. The useful version is not "bigger is better". It is a budget… - [The KV cache and why inference is memory-bound](https://ornix.io/u/omar/large-language-models-end-to-end/kv-cache-memory-bound-inference.md): Training a transformer is limited by compute. Generating from one is limited by memory. If you remember one idea from the inference week, make it this one: at… - [LoRA in one page of math](https://ornix.io/u/omar/large-language-models-end-to-end/lora-in-one-page-of-math.md): Full fine-tuning updates every weight, which means storing gradients and optimizer state for every weight: about 16 bytes per parameter with Adam in mixed… - [RLHF vs DPO](https://ornix.io/u/omar/large-language-models-end-to-end/rlhf-vs-dpo.md): A pretrained model predicts text. An assistant does what you asked. Alignment is the step in between, and for a few years it meant one pipeline: RLHF. DPO… - [Superposition and sparse autoencoders](https://ornix.io/u/omar/large-language-models-end-to-end/superposition-and-sparse-autoencoders.md): A neuron is not a concept. If it were, a layer with 768 neurons could represent 768 things, and language models plainly know far more than that… - [Gradient descent, derived and drawn](https://ornix.io/u/omar/machine-learning-first-principles/gradient-descent-derived-and-drawn.md): Gradient descent is what you do when you cannot solve for the best parameters but you can always tell which way is downhill. Stand somewhere on a bumpy… - [Why cross-entropy and not squared error](https://ornix.io/u/omar/machine-learning-first-principles/why-cross-entropy-not-squared-error.md): The first time you see logistic regression, the loss looks like it fell out of the sky: -[ylog p + (1-y)log(1-p)]. Why not just square the error between the… - [The bias-variance decomposition with real numbers](https://ornix.io/u/omar/machine-learning-first-principles/bias-variance-with-real-numbers.md): Every model makes errors for three different reasons, and only two of them are yours to fix. Bias is the error from a model too simple to express the truth… - [How a decision tree chooses a split](https://ornix.io/u/omar/machine-learning-first-principles/how-a-decision-tree-chooses-a-split.md): A decision tree is twenty questions played greedily. At each node it asks the single yes/no question about one feature that makes the answers on each side as… - [How gradient boosting fits residuals](https://ornix.io/u/omar/machine-learning-first-principles/how-gradient-boosting-fits-residuals.md): A random forest builds many deep trees independently and averages them. Gradient boosting does the opposite: it builds many shallow trees one after another… - [PCA is just an SVD](https://ornix.io/u/omar/machine-learning-first-principles/pca-is-just-an-svd.md): Principal component analysis is sold as a way to "find the directions of greatest variance". That is true, and it hides the fact that you already own the tool… - [Data leakage: five ways to fool yourself](https://ornix.io/u/omar/machine-learning-first-principles/data-leakage-five-ways-to-fool-yourself.md): Data leakage is when information that will not exist at prediction time sneaks into training or evaluation. It is the most common reason a model that scored… - [Reading a confusion matrix and choosing a threshold](https://ornix.io/u/omar/machine-learning-first-principles/reading-a-confusion-matrix-choosing-a-threshold.md): A classifier outputs a score. A decision needs a yes or a no. The step in between, the threshold, is where a model's accuracy turns into real costs and… - [Backprop is the chain rule with bookkeeping](https://ornix.io/u/omar/deep-learning-from-scratch/backprop-chain-rule-bookkeeping.md): A neural network is a big arithmetic expression. Backpropagation is the chain rule applied to that expression in reverse order, with each intermediate value… - [Why initialization matters: the variance math](https://ornix.io/u/omar/deep-learning-from-scratch/why-initialization-matters.md): At initialization a network is a random function. Training can only start from that function, so the first question is whether signals survive the trip… - [Convolution as weight sharing](https://ornix.io/u/omar/deep-learning-from-scratch/convolution-as-weight-sharing.md): A fully connected layer can connect any pixel to any output. Images do not need that. A pixel's meaning depends mostly on its neighbours, and a cat is a cat… - [Residual connections and gradient flow](https://ornix.io/u/omar/deep-learning-from-scratch/residual-connections-gradient-flow.md): In 2015 the ResNet authors reported something odd. A plain 56-layer network had higher training error than a 20-layer one on CIFAR-10. This is not… - [Why LSTMs gate](https://ornix.io/u/omar/deep-learning-from-scratch/why-lstms-gate.md): Recurrent networks reuse one set of weights at every time step, so the gradient from step T back to step 0 passes through the same transformation T times… - [Attention, one head, by hand](https://ornix.io/u/omar/deep-learning-from-scratch/attention-one-head-by-hand.md): Every token in a sequence emits three vectors. A query says what it is looking for, a key says what it contains, and a value says what it will hand over if… - [The ELBO without tears](https://ornix.io/u/omar/deep-learning-from-scratch/elbo-without-tears.md): A VAE is a generative model with a hidden variable z: draw z from a simple prior, then decode it into x. Training it means maximizing log p(x), and that is… - [Diffusion: noise in, image out](https://ornix.io/u/omar/deep-learning-from-scratch/diffusion-noise-in-image-out.md): A diffusion model learns to undo a very small amount of noise, and generates an image by undoing it many times in a row. Take an image and add a little… - [Budgeting tokens and latency](https://ornix.io/u/omar/building-ai-agents/budgeting-tokens-and-latency.md): An agent is a loop, and every turn of the loop re-reads everything that came before. That one fact explains most agent bills and most agent slowness. This… - [The agent loop in 60 lines](https://ornix.io/u/omar/building-ai-agents/the-agent-loop-in-60-lines.md): An agent is a model calling tools in a loop until it decides it is done. That is the whole idea. The loop itself is about 25 lines. Everything else in this… - [Designing tools a model can use](https://ornix.io/u/omar/building-ai-agents/designing-tools-a-model-can-use.md): A tool is a contract between a deterministic program and a model that reads English. Most agent failures I see are tool failures wearing a model's face: the… - [When retrieval beats a longer context](https://ornix.io/u/omar/building-ai-agents/when-retrieval-beats-a-longer-context.md): The question is not "RAG or no RAG". It is "what is the cheapest way to put the right 3,000 tokens in front of the model for this question?" Sometimes that is… - [Context is a budget](https://ornix.io/u/omar/building-ai-agents/context-is-a-budget.md): The context window is not memory. It is a budget that every step spends from, and the model attends to it unevenly. An agent that runs for 40 steps will fill… - [MCP from the wire up](https://ornix.io/u/omar/building-ai-agents/mcp-from-the-wire-up.md): MCP is an open protocol that lets any AI application talk to any tool server through one interface. If you can read JSON-RPC, you can read the whole thing… - [Evals you can trust](https://ornix.io/u/omar/building-ai-agents/evals-you-can-trust.md): "It seems better" is the most expensive sentence in agent engineering. An eval turns it into a number you can defend. But a bad eval is worse than none… - [Prompt injection is a confused-deputy problem](https://ornix.io/u/omar/building-ai-agents/prompt-injection-is-a-confused-deputy-problem.md): Prompt injection is not a prompt problem. It is a confused deputy problem. The agent holds authority: your credentials, your tools, your user's data… ## Docs - [Overview](https://ornix.io/docs.md): What Ornix docs cover and where to start. - [Connect your AI](https://ornix.io/docs/mcp.md): Point Claude, ChatGPT, or any MCP-compatible client at Ornix. - [Tool reference](https://ornix.io/docs/mcp/tools.md): Every tool the connector exposes, with parameters. ## How agents should work with Ornix - Start every session with `ornix_get_context`. It returns the active roadmap, this week's goals and the learning memory brief, so begin from where the learner stopped. - Ask the learner before you change their roadmap, notes or flashcards, and say what you are about to write. - Keep plan edits small and targeted. Use `ornix_shift_schedule` when the learner was away, `ornix_update_week` to rebalance one week, and `ornix_insert_week` or `ornix_remove_week` to add or drop a week. Never regenerate a whole roadmap. - Record learning memory from the learner's own answers, not from guesses. Describe what they did, specifically and kindly. ## Optional - [llms-full.txt](https://ornix.io/llms-full.txt): Every docs page in one Markdown file.