# AI Systems > Practical AI engineering: LLM apps, RAG, chunking, tool calling, agents, evals, and model/transformer foundations. Canonical: https://learn.significanthobbies.com/curriculum/tracks/ai-systems Practical AI engineering: LLM apps, RAG, chunking, tool calling, agents, evals, and model/transformer foundations. This track contains 32 connected concepts. Mastery means explaining each mechanism, predicting its failure modes, and supporting decisions with code, measurements, or a reviewable design artifact. ## Roadmaps - [90-Day AI Search & Infrastructure](https://learn.significanthobbies.com/curriculum/roadmaps/ai-search-infra-90-day) — Build a strong retrieval, vector, AI-systems, and storage foundation through HighSignal and Codevetter. - [12-Month Advanced AI Infrastructure](https://learn.significanthobbies.com/curriculum/roadmaps/ai-infra-12-month) — Reach the depth to design and reason about systems like Turbopuffer, and ship them through HighSignal and Codevetter. - [Runtime — what every runtime has to do](https://learn.significanthobbies.com/curriculum/roadmaps/runtime) — Build one cross-cutting mental model for V8, JVM, Go, BEAM, .NET CLR, vLLM, SGLang, Cloudflare Workers, and wasmtime. - [The Software Engineering Landscape (2026)](https://learn.significanthobbies.com/curriculum/roadmaps/swe-landscape) — Get a working mental model of every major systems-software domain — LLMs, DBs, streaming, game engines, containers, browsers, compilers, OS, networking, distributed, build, crypto. - [12-Week AI Models & Training](https://learn.significanthobbies.com/curriculum/roadmaps/ai-models-training-12w) — Move from transformer foundations through pre-training, fine-tuning, post-training, compression, and evaluation. - [Trace a Tensor](https://learn.significanthobbies.com/curriculum/roadmaps/trace-a-tensor) — Follow one model workload from representation and gradients to memory, kernels, scheduling, and serving economics. ## Concepts - [RAG](https://learn.significanthobbies.com/curriculum/concepts/rag) (core) — Retrieval-Augmented Generation: ground an LLM answer in retrieved context. - [Chunking](https://learn.significanthobbies.com/curriculum/concepts/chunking) (core) — Splitting documents into retrievable units that preserve meaning. - [Context Packing](https://learn.significanthobbies.com/curriculum/concepts/context-packing) (advanced) — Ordering and budgeting retrieved context within the model's window. - [Structured Outputs](https://learn.significanthobbies.com/curriculum/concepts/structured-outputs) (core) — Forcing LLM output into a validated JSON schema. - [Tool Calling](https://learn.significanthobbies.com/curriculum/concepts/tool-calling) (core) — Letting an LLM invoke functions/APIs via structured calls. - [Agent Loops](https://learn.significanthobbies.com/curriculum/concepts/agent-loops) (advanced) — The plan → act → observe loop, with memory and stopping conditions. - [Model Routing](https://learn.significanthobbies.com/curriculum/concepts/model-routing) (core) — Sending each request to the cheapest model that can handle it. - [Prompt & Version Logging](https://learn.significanthobbies.com/curriculum/concepts/prompt-versioning) (core) — Treating prompts as versioned artifacts with logged inputs/outputs. - [LLM Evals](https://learn.significanthobbies.com/curriculum/concepts/llm-evals) (core) — Measuring LLM output quality with datasets, graders, and LLM-as-judge. - [ML Math Foundations](https://learn.significanthobbies.com/curriculum/concepts/ml-math) (intro) — Vectors, matrices, dot products, matmul, shapes. - [Gradient Descent](https://learn.significanthobbies.com/curriculum/concepts/ml-gradient-descent) (core) — Loss surfaces, learning rate, SGD steps. - [Backpropagation](https://learn.significanthobbies.com/curriculum/concepts/ml-backprop) (core) — Chain rule, autograd, forward/backward passes. - [Softmax & Cross-Entropy](https://learn.significanthobbies.com/curriculum/concepts/ml-softmax-xent) (core) — Logits to probabilities, negative log-likelihood loss. - [AdamW Optimizer](https://learn.significanthobbies.com/curriculum/concepts/ml-adamw) (core) — Moments, weight decay, gradient clipping. - [Tokenization (LLM)](https://learn.significanthobbies.com/curriculum/concepts/ml-tokenization) (intro) — Byte-level, char-level, BPE, vocab design. - [Language Modeling](https://learn.significanthobbies.com/curriculum/concepts/ml-language-modeling) (core) — Next-token prediction, context windows, perplexity. - [Sampling & Decoding](https://learn.significanthobbies.com/curriculum/concepts/ml-sampling) (core) — Temperature, top-k, greedy decoding. - [Embeddings (Transformer)](https://learn.significanthobbies.com/curriculum/concepts/ml-embeddings) (core) — Token & position embeddings, tied weights. - [Self-Attention](https://learn.significanthobbies.com/curriculum/concepts/ml-self-attention) (advanced) — Q/K/V, scaled dot-product, causal masking. - [Multi-Head Attention](https://learn.significanthobbies.com/curriculum/concepts/ml-multi-head) (advanced) — Parallel heads, head_dim split, output projection. - [Transformer Block](https://learn.significanthobbies.com/curriculum/concepts/ml-transformer-block) (advanced) — Pre-LayerNorm, residuals, MLP, GELU. - [Training & Debugging](https://learn.significanthobbies.com/curriculum/concepts/ml-training) (advanced) — Init, NaNs, overfit tests, gradient checks. - [Checkpointing](https://learn.significanthobbies.com/curriculum/concepts/ml-checkpointing) (core) — Weights + optimizer state, resume, dataset manifests. - [LoRA & PEFT](https://learn.significanthobbies.com/curriculum/concepts/ml-lora) (advanced) — Frozen base, low-rank adapters, rank/alpha. - [RL Alignment (GRPO & Policy Gradient)](https://learn.significanthobbies.com/curriculum/concepts/ml-rl-alignment) (advanced) — Policy gradient, advantages, GRPO/CISPO, on- vs off-policy RLHF. - [Training Data Engineering](https://learn.significanthobbies.com/curriculum/concepts/ml-data-engineering) (core) — Cleaning, dedup, JSONL tasks, memorization tests. - [Browser ML Runtime](https://learn.significanthobbies.com/curriculum/concepts/ml-browser-runtime) (advanced) — Web Workers, WASM, OPFS, TypedArrays. - [WebGPU Compute](https://learn.significanthobbies.com/curriculum/concepts/ml-webgpu) (advanced) — WGSL, compute kernels, matmul, CPU parity. - [Model Evaluation](https://learn.significanthobbies.com/curriculum/concepts/ml-evaluation) (core) — Held-out loss, baselines, hallucination, leakage. - [Model Pre-training](https://learn.significanthobbies.com/curriculum/concepts/ml-pretraining) (core) — Data mixtures, next-token objectives, scaling laws, distributed training, checkpoints, and training stability. - [Model Quantization](https://learn.significanthobbies.com/curriculum/concepts/model-quantization) (core) — Post-training and quantization-aware methods, integer and low-bit formats, calibration, kernels, and quality trade-offs. - [Open-Weight Models](https://learn.significanthobbies.com/curriculum/concepts/open-weight-models) (core) — Model cards, licenses, weights, tokenizers, chat templates, adapters, provenance, and reproducible packaging.