What is an Algorithm vs. What is Artificial Intelligence?
The ultimate starting point: why traditional code is a rigid recipe written by humans, while machine learning is a computer discovering patterns from thousands of real-world examples.
Artificial Intelligence Academy & Research Hub
XBI, AI Education and Applied Engineering Guides
A multi-page curriculum from first-principles linear algebra, multivariable optimization, and backpropagation to frontier transformer architectures, autonomous agent loops, and enterprise workflow playbooks.
97 Modules
Crawlable Guides
Robotics Lab →
Physical AI & Kinematics
Mechanics Lab →
Softmax, Vectors & Tools
80-Yr History →
Dual AI & Robotics Track
Tool Directory →
40+ Evaluated Engines
Levels 100–700
Academic Tiers
Applied Track
Work, Business & Code
DAG Builder →
Agentic Topology Canvas
WAN Map →
Subsea & Latency Simulator
Pricing Index →
Frontier API vs H100 TCO
SWE-Bench Lab →
Agent Sandboxing & Pass@k
57 Terms →
AI Glossary & Lexicon
Hardware Lab →
LLM VRAM Calculator
Token Lab →
Context & Cost Simulator
Latency Lab →
TTFT & Roofline Simulator
Agentic Lab →
MCP & Topology Simulator
Benchmark Lab →
Capability & Evals Matrix
Energy Lab →
Megawatts & PUE Simulator
Scaling Lab →
Chinchilla & FLOPs Planner
Zero Fabrication
Tested Derivations
How Machines Learn: The Guess, Check, and Nudge Game
The universal mechanism behind all artificial intelligence: how a computer starts with wild guesses, measures its mistakes, and nudges its internal dials closer to the truth.
How Computers 'See' and 'Read': Turning the World into Numbers
Computers cannot see colors or read letters—they only understand numbers. Discover how photos become grids of pixels and how words become coordinates on a giant idea map (embeddings).
What is a Neural Network? (Brains Made of Dominoes and Seesaws)
Demystifying artificial brains: how millions of simple math decision-makers pass clues to one another in layers, moving from simple edges to complex thoughts.
How Large Language Models (LLMs) Work: The World's Smartest Autocomplete
How do tools like ChatGPT, Claude, and Gemini write stories, answer questions, and write code? By playing the world's fastest guessing game: predicting the next word.
AI Safety, Hallucinations & Thinking Like a Scientist
Why does AI sound so confident even when it makes things up? Learn what hallucinations are, why models reflect human biases, and how to use AI as a thinking tool rather than a magic oracle.
Linear Algebra for High-Dimensional Spaces
Vector spaces, linear mappings, dot products as geometric similarity, eigenvalues, and Singular Value Decomposition (SVD) underlying modern embeddings.
Multivariable Calculus & Gradient Optimization
Deriving partial derivatives, gradients, the Jacobian, and Hessian matrices. Visualizing high-dimensional loss landscapes and building gradient descent solvers.
Probability, Distributions & Information Theory
Conditional probability, Bayes' theorem, maximum likelihood estimation (MLE), Shannon entropy, cross-entropy, and Kullback-Leibler (KL) divergence.
Supervised Regression, Classification & Regularization
Ordinary Least Squares (OLS), L1 Lasso and L2 Ridge penalties, logistic odds ratios, and Newton-Raphson numerical solvers.
Generalization Theory & Statistical Validation
The mathematical bias-variance decomposition, cross-validation architectures, hyperparameter tuning, and ROC-AUC curves.
Tree Ensembles & Non-Parametric Models
Information gain, Gini impurity, CART decision trees, Random Forest bagging mechanics, Gradient Boosted Decision Trees (XGBoost), and Kernel SVMs.
The Multilayer Perceptron & Backpropagation
Matrix calculus derivations of backward gradient flow and writing a fully functional neural network in 100 lines of raw Python without PyTorch.
Activation Dynamics & Modern Optimizers
Vanishing/exploding gradients, ReLU vs. GELU vs. SwiGLU gating, momentum, RMSprop, Adam, and AdamW decoupled weight decay mechanics.
Convolutional Networks & Spatial Inductive Biases
Discrete 2D convolutions, receptive fields, pooling layers, residual connections (ResNet), and translation equivariance in computer vision.
The Self-Attention Mechanism Derived
Scaled dot-product attention formulated as content-based associative retrieval in embedding space, temperature scaling, and causal masking.
The Complete Modern Transformer Architecture
Multi-Head Attention (MHA), Grouped-Query Attention (GQA), Rotary Position Embeddings (RoPE), RMSNorm, and pre-layer normalization.
Tokenization & Pretraining Dynamics
Byte-Pair Encoding (BPE), vocabulary compression, token boundary effects, loss curves, compute budgets, and Chinchilla scaling laws.
Supervised Fine-Tuning & Parameter-Efficient LoRA
Full parameter adaptation vs. Low-Rank Adaptation (LoRA), 4-bit NormalFloat quantization (QLoRA), and rank selection algebra.
Preference Alignment: RLHF & Direct Preference Optimization
PPO reinforcement learning from human feedback, Bradley-Terry preference models, Direct Preference Optimization (DPO) derivations.
Test-Time Compute Scaling & Reasoning Models
Inference-time compute frontiers, Tree-of-Thought search, Monte Carlo planning, Process Reward Models (PRMs), and self-correction verification.
Autonomous Agents & Model Context Protocol (MCP)
ReAct execution loops, Model Context Protocol (MCP) standardized transport, tool sandboxing, and deterministic state verification.
Distributed Training Clusters & Hardware Topologies
Distributed Data Parallel (DDP), Fully Sharded Data Parallel (FSDP), Megatron-LM tensor and pipeline parallelism, and InfiniBand interconnects.
Multi-Modal Tokenization: Vision Patches & Unified Latents
Deriving ViT patch linear projections, 2D convolution equivalence, audio mel-spectrogram framing, and continuous cross-attention latent spaces.
Context Caching Economics & KV-Cache Eviction Algorithms
Mathematical models of prefix caching, hash-based PagedAttention block lookup, eviction heuristics, and multi-turn cost reduction.
Open-Weights vs Frontier API Total Cost of Ownership
Deriving hardware capex amortization, datacenter power contracts, support overhead, and exact token crossover inflection points.
Repository-Level Code Context & AST-Driven Retrieval
Replacing naive chunking with structural AST grammars, caller-callee dependency graphs, symbol PageRank, and BM25 + dense hybrid retrieval.
Test-Driven Agentic Code Synthesis & Self-Healing Loops
Eliminating hallucinated patches in autonomous agents: fail-to-pass reproduction, stack trace pruning, and iterative repair convergence.
MicroVM Sandboxing & Secure Tool Execution
Isolating untrusted autonomous agent bash execution at scale: microVM boot latency, memory overhead, seccomp filters, and ephemeral scratch spaces.
The Daily Knowledge-Worker AI Stack
The battle-tested daily operating system for triaging inboxes, drafting high-stakes memos, synthesizing 100-page regulatory PDFs, and accelerating desk output.
Prompting as Structured Thinking
Transition from open-ended chat prompts to deterministic system instructions: Context, Role, Constraints, Few-Shot Demonstrations, and Strict JSON Output Schemas.
Enterprise Support Automation without Hallucinations
Deploying retrieval-augmented generation (RAG) customer support systems with strict fail-closed verification rules so assistants answer strictly from verified documentation.
Personal Document Intelligence with NotebookLM
Grounding personal AI assistants on tax forms, healthcare records, lease agreements, and family schedules to answer complex life questions with verifiable citations.
The Socratic AI Study Partner for Accelerated Learning
Configuring language models as rigorous Socratic tutors that force active recall, pinpoint conceptual misconceptions, and drill complex STEM derivations.
From Plain English to Deployed Web Application
The complete software engineering workflow for solo founders and engineers using Cursor, Claude, and modern deployment pipelines to ship full applications in hours.
The Evolution of Artificial Intelligence: From Symbolic Logic to Reasoning Models
A comprehensive technical history: why symbolic expert systems failed, how GPUs unlocked deep learning, how the Transformer sparked empirical scaling laws, and why the frontier has shifted to test-time compute.
The Silicon & Hardware Behind AI: GPUs, VRAM & The Memory Wall
The microarchitecture of modern machine learning: SIMD execution, Tensor Cores, High-Bandwidth Memory (HBM3e), the Roofline Model, and why memory bandwidth dictates token generation latency.
Byte-Pair Encoding & Tokenizer Mechanics: From Raw Bytes to Subwords
The mathematical bridge from Unicode bytes to semantic subwords: deriving the BPE merge algorithm, vocabulary compression ratios, tokenizer quirks, and multilingual byte inflation.
Prompt Caching & Ephemeral KV Cache Architecture
Prefix matching mechanics in GPU high-bandwidth memory (HBM): skipping the compute-bound prefill phase, cache eviction TTLs, and the prompt prefix hierarchy for 90%+ hit rates.
The Lost in the Middle Phenomenon: Attention Degradation in Long Contexts
Analyzing the U-shaped attention distribution: primacy and recency biases, multi-document retrieval degradation, Needle-In-A-Haystack (NIAH) stress-testing, and hybrid RAG architecture.
Enterprise Token Budgeting & Production Cost Optimization
The production engineering playbook: multi-tier model routing cascades, structured JSON schema token minimization, dynamic sliding-window pruning, and cost-per-outcome observability.
Memory Bandwidth vs. Compute Bound: Arithmetic Intensity & Roofline Analysis
Deriving the hardware roofline boundary: arithmetic intensity (FLOPs/byte), memory bandwidth saturation in auto-regressive decode (GEMV) vs. compute saturation in prompt prefill (GEMM).
FlashAttention & Chunked Prefill: IO-Aware Attention & SRAM Tiling Mechanics
Overcoming the memory wall: FlashAttention-1/2/3 online softmax tiling, SRAM vs. HBM IO complexity reduction from O(N^2) to O(N), and chunked prefill co-scheduling.
Speculative Decoding & Multi-Head Verification: Breaking the Memory Wall
Accelerating auto-regressive decode: small draft model candidate generation, target model parallel verification in a single forward pass, rejection sampling proofs, and Medusa speculative heads.
Continuous Batching & PagedAttention: Production LLM Serving Architecture
The production serving stack: eliminating static padding bubbles via iteration-level scheduling, virtual memory block tables for dynamic KV cache allocation (vLLM), and prefill-decode disaggregation.
Model Context Protocol (MCP): JSON-RPC 2.0 & Schema Architecture
Deconstructing the open connectivity protocol: JSON-RPC 2.0 framing, tool discovery (tools/list), invocation (tools/call), resource subscriptions, and SSE vs. stdio transports.
Multi-Agent Topologies: ReAct, Orchestration & Swarm Consensus
Comparative graph topologies: single-loop ReAct, centralized orchestrator-workers, decentralized peer swarms, DAG execution pipelines, and consensus voting mechanisms.
Deterministic Agent Guardrails: State Machines & Human Interception
Enforcing non-bypassable constraints: finite state machines (FSM), JSON schema validation gates, blast-radius containment, tripwires, and human-in-the-loop (HITL) checkpoints.
Production Agent State: Thread Checkpointing & Memory Stores
Architecting robust production storage: transactional thread checkpoints (SQLite/Postgres), time-travel debugging, semantic vector stores, and episodic vs. working memory.
Evaluation Methodologies: SWE-bench, GPQA & LLM-as-a-Judge
The rigorous science of model evaluation: pass@1 execution harnesses, Docker sandbox execution for SWE-bench Verified, and statistical agreement in LLM-as-a-judge evaluators.
Test-Time Compute Scaling: PRMs & Monte Carlo Search Trees
Deriving test-time scaling laws: replacing raw pretraining compute with Process Reward Models (PRMs), outcome verifiers, and beam search during generation.
Benchmark Contamination: Detection, N-Grams & Synthetic Canaries
Auditing training corpora for test set leaks: 13-gram exact matches, embedding distance anomalies, synthetic Canary GUID validation, and temporal holdouts.
Enterprise Evaluation Systems: Red-Teaming & Regression Suites
Deploying enterprise continuous evaluation: automated jailbreak red-teaming, CI/CD regression gates, latency SLAs, and golden customer dataset tracking.
Megawatt Power Delivery: 115kV Utility Grid to 54V Server Busbars
High-voltage transmission stepdown, transformer topologies, harmonic distortion, UPS battery reserves, and 54V/48V busbar loss mitigation in gigawatt-scale AI datacenters.
Thermal Density & Liquid Cooling: Cold Plates, CDUs & Immersion
Thermodynamics of 1,000W+ processors, microchannel cold plates, primary vs secondary loops, Coolant Distribution Units (CDUs), and dielectric immersion physics.
High-Radix Optical Fabrics: InfiniBand, RoCE & Dragonfly Topologies
Architecture of 800G/1.6T network fabrics: non-blocking Fat-Tree vs Dragonfly+, optical transceivers, Co-Packaged Optics (CPO), and collective communication scaling.
Sovereign AI Infrastructure: Site Selection & SMR Nuclear Contracts
Transmission interconnection queues, dual-feed substation tariffs, Power Purchase Agreements (PPAs), Behind-the-Meter (BTM) generation, and Small Modular Reactor (SMR) economics.
Enterprise Multi-Model Routing & Gateway Architectures
Architecting multi-tier model gateways: classification routing, structured JSON parsing, fallback cascade pools, and 99.99% uptime circuit breakers.
Continuous Delivery with Autonomous Coding Agents
Integrating coding agents into enterprise CI/CD pipelines: webhook issue triage, microVM testing, automated pull requests, and human-in-the-loop gates.
State Graph Checkpointing, Resumption & Memory Compaction
Deterministic state serialization, thread forks, durable execution, append-only logs, and sliding-window KV compaction in multi-agent runtimes.
Deterministic Guardrails & Human-in-the-Loop Hooks
Fail-closed policy enforcers, AST safety filters, schema validators, human approval suspension, and compensation transactions.
Multi-Agent Swarm Consensus: Raft & Byzantine Fault Tolerance
Quorum formation, leader election, Byzantine fault tolerance, log replication, and voting invariants across autonomous swarms.
Subsea Cable Physics: Chromatic Dispersion & Terabit Capacity
Speed of light in silica, erbium-doped fiber amplifiers (EDFA), DWDM spectral efficiency, and transoceanic latency floors.
Distributed Pipeline Parallelism Across High-Latency WANs
1F1B schedules, micro-batch bubble minimization, gradient compression, and geo-distributed tensor execution over 70ms+ round trips.
Sovereign Data Residency & Confidential Cryptographic Enclaves
AMD SEV-SNP, Intel TDX, hardware root of trust, memory encryption keys, and cross-border regulatory compliance in sovereign AI.
Production Agentic Observability & OpenTelemetry Tracing
W3C trace contexts, distributed span hierarchies, token burn telemetry, tool call attribution, and latency bottleneck tracing.
Global Edge Inference Routing & Dynamic KV-Cache Replication
Anycast BGP DNS, edge prefill caching, prompt prefix replication, and cross-continental failover routing for frontier LLM clusters.
AI Glossary & Terminology Lexicon
Master dictionary for all 31 academy courses. Every term features dual-level definitions: "The Simple Idea" (first-principles intuition) and "Engineering Specification" (formulas, tensor shapes, and hardware limits), plus instant live search.
Frontier AI Hardware & Local LLM VRAM Calculator
Interactive systems engineering calculator. Compute exact weights, GQA/MLA KV cache, activations, and memory-bandwidth bound generation speeds (tokens/sec) across consumer RTX GPUs, Apple Silicon unified memory, and datacenter clusters.
Frontier LLM Token Economics & Context Architecture Lab
Interactive systems pricing and context budget simulator. Calculate multi-model API costs, prompt caching return on investment (ROI), 6-part context window allocation, and enterprise monthly run-rates across Claude 3.7, GPT-4.5, DeepSeek-R1, and Gemini.
Inference Latency, Throughput & TTFT Systems Lab
Hardware roofline and serving simulator. Compute Time-to-First-Token (TTFT), Inter-Token Latency (ITL), single-stream vs. batched tokens/sec, and speculative decoding speedup across NVIDIA H100, B200, RTX 4090, and Apple M4 Max.
Agentic Workflow & MCP Architecture Lab
Interactive multi-agent systems simulator. Model ReAct, Orchestrator-Workers, and Swarm topologies, inspect Model Context Protocol (MCP) JSON-RPC 2.0 payloads, and calculate quadratic context token burn across multi-turn reasoning loops.
Frontier AI Benchmark & Capability Lab
Interactive systems evaluation and capability matrix. Compare SWE-bench Verified, GPQA Diamond, LiveCodeBench, Math 500, Pass@k sampling curves, and cost-per-accuracy economics across Claude 3.7 Sonnet, OpenAI o3-mini, and DeepSeek-R1.
Frontier Model Pricing Index & Multi-Modal Token Engine
Comprehensive pricing index and TCO calculator. Compare input/output token costs across Claude 3.7, GPT-4.5, Gemini 2.5, DeepSeek-R1, and Qwen 2.5. Calculate prompt caching discounts, multi-modal tokens (images, audio, video), and 8x H100 self-hosted GPU breakeven volume.
Autonomous Coding Agent & SWE-Bench Execution Lab
Interactive autonomous software engineering simulator. Ingest repository AST context, simulate multi-turn self-healing test repair loops, benchmark microVM sandboxing overhead (Firecracker vs gVisor vs Docker), and plot pass@k statistical convergence.
Autonomous Agent Visual DAG & Workflow Builder
Interactive node-graph topology canvas for autonomous multi-agent systems. Wire orchestrators, deterministic guardrails, subagents, and human-in-the-loop gates. Simulate context accumulation and export production-ready code.
Global AI Compute & Subsea Interconnect Latency Map
Interactive planetary interconnect and transoceanic fiber routing simulator. Model speed-of-light propagation in silica (200,000 km/s), chromatic dispersion, and determine feasibility of distributed pipeline parallelism across continents.
Datacenter Energy & Sovereign AI Infrastructure Simulator
Model cluster megawatt power delivery from 115kV transmission lines to 54V server busbars. Calculate Power Usage Effectiveness (PUE), liquid CDU flow rates, direct-to-chip cooling thermal resistance, and SMR nuclear baseload requirements.
Frontier AI Scaling Laws & Pre-Training Compute Budget Planner
Calculate exact 6ND pretraining compute FLOPs, Chinchilla-optimal parameter/token balance, MoE active parameter throughput, cluster training duration, and CapEx costs for GB200, B200, H100, and TPU v5p clusters.