un

სავაჭრო უნარები. რეალური კარიერა. შენი გზა.

Grid Grouped Catalog

გაკვეთილის შექმნა

შედით სისტემაში, რომ შექმნათ და გააზიაროთ საკუთარი გაკვეთილები.

Orbital Mechanics: Getting to Space

From Newton's cannonball to Hohmann transfers: the physics that gets rockets into orbit and keeps satellites from falling.

agriculture

Agriculture: Growing Food

Soil science, crop management & sustainable farming: knowledge that feeds the world & the careers that depend on it.

agriculture

Geometry of Agriculture

Center pivot circles, contour farming, row spacing, drainage slope & field layout: geometry that feeds the world.

automotive

Automotive: How Cars Work

From internal combustion to EVs: understand engines, drivetrains, and diagnostics, the foundation of every auto tech career.

aviation

Aviation: Principles of Flight

Lift, drag, thrust & weight: four forces that get aircraft airborne, & the careers that keep them flying.

aviation

Geometry of Aviation

Flight vectors, bank angles, turn radius, wind triangles, and approach geometry: the math pilots use every day.

business

Financial capital represents one of eight. Living, material, intellectual, experiential, social, cultural & spiritual capital each compound differently; each can grow, deplete, or exchange. Learn to see all of them.

business

Entrepreneurship: Starting a Business

From idea to revenue: learn business models, market validation, and the fundamentals every founder needs before launch day.

business

Geometry of Collaboration

Triangulation, Voronoi domains & tessellation: geometric principles behind the machine learning triangle. Two solos with complementary specifications, one machine learning bridge, one gap to close.

business

ML Triangle: Beyond Unturf

Unturf represents one instance of the machine learning triangle pattern, not the only one. Two solos with complementary specifications plus one machine learning engine plus one gap equals a bridge, across any of the eight forms of capital. Learn to recognize the pattern, find a complementary solo & write a specification that makes the triangle form.

business

What Does Kanban Mean?

From Toyota factory floors to cross-functional work centers: learn a pull system, the board, WIP limits & how to coordinate work across a product studio, bakery, construction site, or any multi-discipline operation without meetings or managers.

cnc machining

CNC Machining: Precision Manufacturing

From G-code to finished parts: learn how computer-controlled machines shape metal, plastic, and composites with thousandths-of-an-inch precision.

cnc machining

Geometry of CNC Machining

Coordinate systems, tool paths, G-code arcs, work envelopes, and GD&T: the geometry that makes precision manufacturing possible.

communication

A Diagram Is Worth 3,000 Words

Visual communication gives you a professional superpower. This lesson covers why diagrams outperform prose, major diagram types, how to read complex visuals systematically & how to design diagrams that actually communicate.

computer science

Build a Stemmer: Harris 1956 and the Roots of NLP

Implement an English stemmer from scratch: learn morphological analysis, suffix stripping, and test-driven development.

computer science

Capacity Planning: Art & Science

John Allspaw's discipline of running services that scale without surprise: measurement, forecasting, ceiling tests, headroom buffers, & the diagonal scaling decision.

computer science

Cybersecurity: Defense in Depth

Think like a defender: learn threat modeling, network security, and incident response from the ground up.

computer science

Data Science: From Spreadsheets to Models

Bridge the gap from Excel to Python: learn how data scientists clean, explore, and model real-world data.

computer science

Geometry in Computer Science and Machine Learning

How geometric thinking powers everything from nearest-neighbor search to neural network training: vectors, distances, manifolds, and loss landscapes.

computer science

Geometry of Capacity Planning

Visual structure underneath capacity work: Little's Law as area, queueing curves with their elbow at 80%, trend extrapolation as slope, & headroom envelopes as 2D safety margins.

computer science

Geometry of Cybersecurity

Network topology, attack surface geometry, graph theory in threat modeling, and the geometric structure of encryption.

computer science

Geometry of Site Reliability Engineering

The visual structure underneath SRE: latency distributions as long-tail curves, error budgets as 2D cones, dependency graphs as DAGs, & dashboard layouts as information geometry.

computer science

Site Reliability Engineering

What Google invented when ops became too important to leave to manual labor. SLOs, error budgets, toil elimination, blameless postmortems, & the four golden signals.

construction

Construction: Building from the Ground Up

From foundations to framing to finish: understand how buildings come together and the trades that make it happen.

construction

Geometry of Construction

Roof pitch, rafter angles, stair geometry, arch types, and site layout: the geometry carpenters and builders use every day.

cosmetology

Geometry of Haircutting

The head as a sphere, elevation and projection angles, graduation wedges, traveling guides, and the balanced silhouette: precision geometry behind a great cut.

cs algorithms

Algorithms in Practice: Code Review & Complexity Defects

Professional software development requires reading code for algorithmic complexity, not just correctness. This lesson covers complexity-aware code review, real patterns found in production systems, interview-style analysis, & a MOAD-0001 defect found in 1,000+ open-source projects — with a direct pointer to the unhamming course for the full investigation.

culinary

Culinary Arts: Kitchen Fundamentals

Knife skills, flavor profiles & kitchen management: foundation every professional cook needs before the first ticket fires.

culinary

Geometry of Culinary Arts

Knife cut geometry, plating composition, baking ratios, and pan sizing: precision geometry in the kitchen.

education

Geometry of Facilitation

Vision cones and occlusion, the room as a graph and the sweep as a watchman route, triangulating where help is needed, and the proxemics of intervention distance: the geometry a floating teacher works in.

education

The Captain Faces the Crew: Running a Self-Directed Learning Studio

A practical lesson for parents, pod facilitators, and trade-school instructors: how a floating teacher runs a room of self-paced learners: daily rhythm, where your attention goes, and assessment without grading.

electrical

Breaker Boxes and Fuse Boxes: Overcurrent Protection

From service entrance to individual circuits: how breaker panels, sub-panels, and fuse boxes protect your wiring, and what every electrician needs to read one correctly.

electrical

Electrical Wiring: Residential Circuits

Understand residential electrical systems: from breaker panels to outlet wiring, US and UK residential wiring compared, the knowledge every electrician starts with.

electrical

Electrical: Industrial Systems and Datacenters

Three-phase power, motor control centers, VFDs, and how datacenters keep the lights on: industrial electrical from switchgear to server rack.

electrical

Geometry of Electrical Work

Conduit bending math, box fill calculations, wire routing geometry, and the electromagnetic fields that follow geometric laws.

electrical

Running 110V: Outlets, Switches, and Light Fixtures

From pulling cable to making connections: wire an outlet, a switch loop, and a light fixture correctly the first time. NEC code, box fill, and wire termination technique.

entrepreneurship

unturf Partnership Program

unturf holds machine-learning-first ventures as an umbrella organization. This lesson covers the structure, permacomputer economy, extreme leverage, ML Triangle, and how external partners join: no change required to your existing enterprise.

environmental

Renewable Energy: Powering the Future

Solar, wind, hydro & storage: understand the energy transition & booming careers in clean power.

finance

Personal Finance: Money Fundamentals

Budgeting, investing, and avoiding debt traps: the financial literacy every adult needs and nobody teaches in school.

geometry of cs algorithms

Geometry of CS Algorithms

Every algorithm has a shape. O(1) draws a flat horizon. O(N) draws a diagonal. O(N²) draws a parabola. Binary search halves a line segment. A hash table maps a key to a bucket like a function maps a domain to a range. Sorting networks visualize as parallel comparison wires. This lesson reads algorithms geometrically.

geometry of machine learning

Geometry of PAC Learning

Every chapter of PAC learning carries a shape. ε & δ draw a budget plane. VC shattering draws dichotomies on point clouds. PAC-Bayes draws a posterior cloud over hypothesis space. Double descent draws a curve that classical theory predicted should rise but actually falls. Scaling laws draw a 3D power-law surface. Empirical audit draws a Beta posterior tightening into a needle. Six chapters, six geometries, one activity.

geometry of machine learning

Geometry of ANDREA: Five Spatial Pictures of Language Model Training

Every part of ANDREA carries a shape. Embedding space draws semantic distance as Euclidean distance in R^768. Attention draws a projection of past tokens onto a query subspace. Loss surface draws a high-dimensional terrain that gradient descent walks. Bandit weights draw a discrete simplex where each vertex is one data source & every interior point is a curriculum mix. Curriculum warmup draws a restricted submanifold projection. Five geometries, one model. Walk each picture & lock in the spatial intuition that abstract equations alone cannot deliver.

geometry of unhamming

Geometry of Unhamming

Hamming saw geometry everywhere: n-dimensional space, sphere packing in error correction, the collapse of high-dimensional intuition. Unhamming extends this: O(N²) as area, O(N) as a line, the factory model as a DAG, & complexity classes as geometric shapes. Every algorithm has a shape. Learning to see the shape predicts the cost.

hamming on hamming

Hamming on Hamming

Richard Hamming's graduate lectures on style of thinking in science & engineering: knowledge explosion, creativity through analogy, expert blindness, data reliability, measurement distortion, & the habits separating people who do important work from those who stay busy.

hamming on hamming

Geometry of Hamming

The geometric structures underlying Hamming's work: Hamming distance in binary n-space, error correction via sphere packing, the √n vs n random walk argument, & exponential growth models for knowledge doubling & half-life of expertise.

hamming on hamming

Foundations of the Digital Revolution

Hamming's nine-point argument for why digital beats analog: reliability of discrete values, IC interconnection economics, the shift from material to information economy, robots & manufacturing, the equivalent product insight, & the military information dominance thesis.

hamming on hamming

Geometry of the Digital Revolution

The geometric structures behind Hamming's Chapter 2: logarithmic cost curves & orders of magnitude, signal-to-noise ratio geometry in analog vs digital channels, & exponential models for technology adoption — the S-curve as a logistic equation.

hamming on hamming

History of Computer Hardware

Hamming's survey of hardware evolution: Napier bones through ENIAC & IBM 701, the exponential speed curve from relay to VLSI, three physical limits (molecular size, speed of light, heat dissipation), & the lesson experts who predicted only 3-5 computers would ever be needed failed to learn.

hamming on hamming

Geometry of Computer Hardware

The geometric structures behind Hamming's Chapter 3: exponential growth & physical ceilings on a log-log plot, the speed-of-light constraint as a sphere of influence, heat dissipation as a surface-area-to-volume problem, & parallel speedup via Amdahl's Law.

hamming on hamming

History of Computer Software

Hamming's survey of software evolution: from absolute binary to FORTRAN & ALGOL, spaghetti code & relocatable libraries, the psychological vs logical language design distinction, & four criteria for successful programming languages that predict which survive 60 years.

hamming on hamming

Geometry of Computer Software

The geometric structures behind Hamming's Chapter 4: parse trees & formal grammars as directed acyclic graphs, information-theoretic redundancy in language design, complexity classes (Big-O) as geometric growth curves, & the language adoption curve as a network-effect model.

hamming on hamming

History of Computer Applications

Hamming's survey of how computing moved from number-crunching to economics-driven applications: the S-curve of technology adoption, the equivalent product insight applied to computerization, the Boeing shared design tape story, pattern recognition as the next frontier, & the career lesson from giving the same talk for 30 years.

hamming on hamming

Geometry of Computer Applications

The geometric structures behind Hamming's Chapter 5: the logistic S-curve as a differential equation, network-effect growth as a quadratic utility function (Metcalfe's Law), & optimization landscapes — gradient descent, local minima, & why shared mutable state produces phantom gradients.

hamming on hamming

Limits of Machine Reasoning: I

Hamming's examination of the fundamental question: can machines think? The Turing test, tacit knowledge, game-playing programs, expert systems, & the boundary between rule-based logic & human judgment.

hamming on hamming

Geometry of Machine Reasoning: I

Search tree geometry: branching factor, depth, exponential node counts, alpha-beta pruning reduction, & the geometric structure that makes rule-based game programs tractable.

hamming on hamming

Limits of Machine Reasoning: II

Hamming on machine creativity: the geometry theorem-proving program & the isosceles triangle proof it discovered, computer music at Bell Labs, the automation-displacement question, & the human-machine collaboration frontier.

hamming on hamming

Geometry of Machine Reasoning: II

Formal proof as a path in axiom space; projective duality & the selfcongruence proof of the isosceles triangle theorem; Nyquist sampling geometry; proof-length minimization as a search problem.

hamming on hamming

Limits of Machine Reasoning: III

Hamming's rebuttals to common objections about machine capability, the question of career relevance, & the transition to n-dimensional design space — bridging machine reasoning limits to the geometry of high-dimensional problems.

hamming on hamming

Geometry of Machine Reasoning: III

Decision boundary geometry: linear separability, the XOR non-separability problem, the perceptron as a hyperplane classifier, feature space transformations that enable separation, & the kernel trick as a geometric embedding.

hamming on hamming

Hamming Ch. 9: n-Dimensional Space

Hamming's treatment of n-dimensional geometry: Stirling's approximation, the Gamma function, the volume of n-dimensional spheres, & the counterintuitive collapse of high-dimensional geometry that governs all engineering design with many free parameters.

hamming on hamming

Geometry of n-Dimensional Space

Stirling's approximation as a geometric tool, the n-sphere volume formula with numeric verification, the corner paradox computed explicitly, & the implications of high-dimensional geometry for optimization & machine learning.

hamming on hamming

Hamming Ch. 10: Coding Theory I

Shannon's communication model, the two-stage encoding architecture (source coding & channel coding), variable-length codes, the Kraft inequality, Shannon entropy, the noiseless coding theorem, & prefix-free decoding trees.

hamming on hamming

Geometry of Coding Theory I

The binary tree geometry of prefix-free codes, Kraft's inequality as a geometric constraint on tree structure, Shannon entropy as a lower bound on average code length, & numeric verification with a worked example.

hamming on hamming

Hamming Ch. 11: Coding Theory II

Huffman coding algorithm in full detail: the merge-and-split procedure, minimum average length L, code non-uniqueness, bushy vs long trees, variance of code lengths, block codes vs variable-length codes, & self-compiling programs.

hamming on hamming

Geometry of Coding Theory II

Huffman coding as greedy tree construction: expected code length as a weighted sum, variance of code lengths, the bushy vs long tree trade-off, & a numeric worked example computing optimal Huffman codes & verifying the noiseless coding theorem.

hamming on hamming

Error Correcting Codes

Hamming's Chapter 12: the discovery story of error-correcting codes, from rectangular parity checks to the (7,4) Hamming code, syndrome decoding, SECDED, & the sphere-packing interpretation of error correction.

hamming on hamming

Geometry of Error Correcting Codes

The linear-algebraic geometry behind error-correcting codes: parity check matrices as linear maps, syndrome decoding as a quotient-space operation, coset leaders & standard arrays, & the (7,4) Hamming code as a perfect tiling of binary 7-space.

hamming on hamming

Information Theory

Hamming's Chapter 13: Shannon's definition of information & entropy, Gibbs' inequality & the noiseless coding theorem, binary symmetric channel capacity, & the fundamental noisy coding theorem — what it proves, what it costs, & why error-correcting codes complement rather than replace it.

hamming on hamming

Geometry of Information Theory

The geometric structures underlying Shannon's information theory: entropy as a concave function on the probability simplex, channel capacity as a maximum over input distributions, KL divergence as a geometric distance in probability space, & rate-distortion geometry.

hamming on hamming

Digital Filters I: Eigenfunctions & Sampling

Hamming's introduction to digital filters: why complex exponentials serve as eigenfunctions of linear time-invariant systems, the transfer function as eigenvalue, & the Nyquist sampling theorem as the geometric foundation of all digital signal processing.

hamming on hamming

Geometry of Digital Filters I

Frequency as angle on the unit circle: complex exponentials trace circular paths, sampling maps to rotations, & the Nyquist limit emerges from the geometry of half a revolution. Aliasing as wraparound on the circle.

hamming on hamming

Digital Filters II: Design by Conditions & the Gibbs Phenomenon

Designing a filter to meet exact frequency conditions: deriving the 3-sample averaging filter, the transfer function formula, & the Gibbs phenomenon that limits any truncated Fourier series design — with Hamming's account of how Gibbs earned his discovery.

hamming on hamming

Geometry of Digital Filters II

The Z-plane as geometric design space: pole-zero diagrams encode frequency response as distances from the unit circle, zeros cancel frequencies, poles boost them, & stability reads directly from pole positions relative to the unit circle boundary.

hamming on hamming

Digital Filters III: Fourier Design & Kaiser's Method

The systematic Fourier series approach to filter design: start with an ideal response, compute Fourier coefficients, truncate, apply a window to suppress Gibbs ripple, & use Kaiser's method to choose both N and the window from ripple & transition-width specifications.

hamming on hamming

Geometry of Digital Filters III

Fourier series as orthogonal projection: basis functions span a function space, Fourier coefficients are inner products, truncation equals projection onto a subspace, & the Gibbs phenomenon emerges from sinc convolution at jump discontinuities.

hamming on hamming

Digital Filters IV: Recursive Filters & Feedback

IIR (recursive) filters: feedback enables sharp frequency response with fewer coefficients, stability requires poles inside the unit circle, & the classic analog filter families (Butterworth, Chebyshev, elliptic) transformed to discrete time — with Hamming's shower story as the concrete feedback instability example.

hamming on hamming

Geometry of Digital Filters IV

IIR filter geometry: the Z-transform as a map from time to the complex frequency plane, stability as pole confinement inside the unit circle, Butterworth poles on a circle, elliptic poles on an ellipse, & the bilinear transform as a conformal map from the Laplace s-plane.

hamming on hamming

Simulation I: When Experiments Are Impossible

Hamming's first large-scale simulation — Los Alamos atomic bomb calculations — reveals the core principles: expert domain knowledge, repetitive structure, equivalent results, stability vs instability, & the danger of mistaking a simulation for reality.

hamming on hamming

Geometry of Simulation I: Euler's Method & the Step-Size Region

The geometric structures underlying numerical simulation: Euler's method as forward difference in the (x,y) plane, step size h as a geometric parameter with a stability region in the complex plane, discrete dynamical systems, & the simulation feedback loop as iteration.

hamming on hamming

Simulation II: Why Should Anyone Believe It?

Hamming's central question for every simulation: why should anyone believe it? The validity problem, analog vs digital tradeoffs, Simpson's paradox in simulation data, self-reinforcing models, & the double-blind lesson for honest simulation design.

hamming on hamming

Geometry of Simulation II: Model Validity as Geometric Fitting

Simulation validity as geometry: fitting a surface to data points, residuals as distances, systematic bias as offset vectors, the Hawthorne effect as a hidden dimension, & Hamming's validation checklist reframed as geometric constraints on model fit.

hamming on hamming

Simulation III: GIGO, Stability, & Solving the Unsolvable

Hamming challenges the GIGO assumption: garbage-in does not always mean garbage-out. Feedback within a computation can compensate for imprecise inputs. Numerical ODE solvers as recursive digital filters, predictor-corrector methods, the Nike missile instability, & the transistor equation Hamming solved by exploiting its instability.

hamming on hamming

Geometry of Simulation III: Stability Regions, Stiffness, & High-Dimensional Tubes

The full geometric picture of ODE solver stability: Euler & RK4 stability regions in the complex hλ-plane, stiffness as eigenvalue geometry, high-dimensional tube paradox from Chapter 9, & using instability as a navigation signal.

hamming on hamming

Hamming Ch. 21: Fiber Optics

Hamming's account of fiber optic technology: total internal reflection, critical angle, small-diameter advantages, electromagnetic immunity, & the security insight Hamming recognized before anyone asked for it.

hamming on hamming

Geometry of Fiber Optics

The geometric structures underlying fiber optics: Snell's law as angle transformation, critical angle as a geometric threshold, numerical aperture as acceptance cone geometry, evanescent wave decay, & single-mode diameter as a constraint on propagation angle space.

hamming on hamming

Hamming Ch. 22: Computer-Aided Instruction

Hamming's analysis of computer-aided instruction (CAI): the Hawthorne effect & its corrosive impact on educational experiments, the double-blind problem in pedagogy, branching programs, immediate feedback, & the question of whether machines can genuinely improve learning.

hamming on hamming

Geometry of Computer-Aided Instruction

The geometric structures underlying learning & memory: power law learning curves, exponential forgetting curves, geometric spacing as optimization, & branching programs as directed graphs — the mathematics that makes adaptive instruction tractable.

hamming on hamming

Hamming Ch. 23: Mathematics

Hamming's exploration of what mathematics is: the Platonic vs Formalist schools, Wigner's 'unreasonable effectiveness,' the universality of mathematics across civilizations, & why abstraction increases applicability.

hamming on hamming

Geometry of Mathematics

The geometric structures underlying mathematics itself: axiomatic systems as formal spaces, proof as a path problem, Gödel incompleteness as a topological property, completeness as connectivity, & the Platonic 'space' of mathematical objects.

hamming on hamming

Hamming Ch. 24: Quantum Mechanics

Hamming's account of quantum mechanics: Heisenberg's matrix mechanics, Schrödinger's wave mechanics, their mathematical equivalence via Hilbert space, the Born rule, & Hamming's act-as-if principle for using formalism without requiring philosophical satisfaction.

hamming on hamming

Geometry of Quantum Mechanics

The geometric structures underlying quantum mechanics: Hilbert space as infinite-dimensional inner product space, quantum state as unit vector, measurement as projection, Born rule as squared projection length, Bloch sphere for qubit geometry, & entanglement as non-separability in tensor product space.

hamming on hamming

Hamming Ch 25: Creativity

Hamming's deep analysis of creativity: how novelty differs from originality, why science fails to recognize creative work when it occurs, the role of the prepared mind, the subconscious gestation process, & the 10 important problems technique.

hamming on hamming

Geometry of Creativity

Creativity modeled as search in combinatorial space: analogical reasoning as graph isomorphism, the prepared mind as a richly connected knowledge graph, serendipity as short paths between distant nodes, & the combinatorial explosion of idea space.

hamming on hamming

Hamming Ch 26: Experts

Hamming on expert resistance: how paradigms filter evidence, the two failure modes of expertise, why great innovations come from outside the field, how to handle expert pronouncements of impossibility, & what to do when you face a change in the dominant paradigm.

hamming on hamming

Geometry of Experts

Paradigm space as a loss landscape with local minima; paradigm shifts as escaping local minima; expert knowledge as deep gradient investment; the topology of knowledge with multiple local minima; Kuhn cycles as gradient descent dynamics.

hamming on hamming

Hamming Ch 27: Unreliable Data

Hamming on measurement: why data is less reliable than advertised, the systematic-vs-random error distinction, calibration chains, the life-testing dilemma, why data fitting theory too well is suspicious, & practical rules for handling data in the field.

hamming on hamming

Geometry of Unreliable Data

Error geometry: measurement uncertainty as probability distributions, systematic error as bias (shift of mean), random error as variance, propagation of uncertainty through calculations via partial derivatives, chi-squared goodness of fit, & suspiciously clean data.

hamming on hamming

Hamming Ch 28: Systems Engineering

Hamming's systems engineering principles: why optimizing components ruins system performance, the importance of interfaces over components, bounding conditions, the education system as a failed systems engineering case, & how to think as a systems engineer rather than a component optimizer.

hamming on hamming

Geometry of Systems Engineering

Production possibility frontiers & Pareto frontiers; why component optima lie off the system optimum; constraint geometry and feasible regions; the maximum principle; interface as a shared boundary in parameter space; why system optima live at constraint boundaries.

hamming on hamming

Hamming Ch 29: You Get What You Measure

Hamming on measurement distortion: how IQ calibration creates the distribution it claims to discover, Goodhart's law before Goodhart named it, the corruption of metrics by targeting, the dynamic range problem in rating scales, & practical defenses against measurement distortion.

hamming on hamming

Geometry of You Get What You Measure

Metric distortion as a mapping from value space to measurement space; Goodhart's law as gradient ascent divergence; multi-objective optimization (Pareto) as defense against single-metric corruption; measurement invariance and why it matters.

hamming on hamming

Hamming Ch 30: You and Your Research

Hamming's direct advice on doing great work: the 10 important problems technique, courage to work on hard problems, the open door policy, compound knowledge growth, the role of luck vs preparation, & the central question every researcher must answer.

hamming on hamming

Geometry of You and Your Research

Research portfolio as a point in problem-space; the 10 important problems technique as maintaining frontier coverage; compound interest of knowledge as convex growth; opportunity cost geometry; why avoiding hard problems means missing high-leverage regions of problem-space.

health

EMT Basics: Emergency Medicine Fundamentals

First link in the chain of survival: learn patient assessment, airway management & emergency protocols.

hvac

Geometry of HVAC

Duct sizing, airflow geometry, coil surface area, psychrometric charts, and the geometric principles of heating and cooling.

hvac

HVAC: Heating and Cooling Systems

From refrigeration cycles to ductwork: understand systems that keep buildings comfortable & the career that keeps them running.

logistics

Geometry of Logistics

Warehouse layout optimization, pallet stacking geometry, route optimization, and the traveling salesman problem.

logistics

Logistics: Supply Chain Fundamentals

How goods move from factory to doorstep: warehousing, routing & the invisible network that keeps the world running.

machine learning

PAC Learning: From Theory to Production Audit

Probably Approximately Correct (Valiant, 1984) gave a first rigorous answer to 'what does it mean for a machine to learn?' This lesson walks all six chapters: classical PAC, VC dimension, agnostic + PAC-Bayes variants, why deep learning broke classical bounds, scaling laws & data-wall estimates for automated general intelligence, & a production audit pattern that makes empirical PAC operational. Six chapters, one activity.

machine learning

Grow a Language Model: One GPU, One Model

ANDREA grew on a single RTX 4090. This first activity walks through what a language model predicts, why a decoder-only transformer fits 12M to 480M parameters on one consumer GPU, & maps 24 follow-up activities covering every algorithm in a recipe (Autonomous Neural Data Recipe for Education & Agency). Start here.

machine learning

Grow a Language Model: Harris Morpheme Tokenizer

ANDREA does not use BPE. It segments words at morpheme boundaries discovered through distributional analysis, building a vocabulary of (256 + N + 1) tokens. This activity walks through what a tokenizer does, how Harris finds morphemes, & why the 256+N+1 structure matters.

machine learning

Grow a Language Model: Tokenizer Diet

ANDREA-12M wasted 63.6% of its vocabulary on data a model never saw. ANDREA-120M fixed that. This activity covers tokenizer-diet alignment, vocabulary saturation, & a vocab-science discovery that 8192 segments saturate a 1.25B-character corpus while 16,384 produces only 13,106 actual segments.

machine learning

Grow a Language Model: Embeddings & Position

Token IDs flow into a transformer through two lookup tables: token embeddings (V × d_model) & learned position embeddings (T × d_model). This activity covers vector spaces, why dot products measure similarity, & why ANDREA uses learned position embeddings instead of RoPE or sinusoidal.

machine learning

Grow a Language Model: Scaled Dot-Product Attention

Attention computes Q · K^T / sqrt(d_k), masks future positions, applies softmax, & weights V. This activity covers each piece, why a sqrt(d_k) factor matters, why a causal mask makes autoregressive generation possible, & how ANDREA-120M splits 768-dimensional input across 12 heads of 64 dimensions each.

machine learning

Grow a Language Model: Multi-Head Attention

One attention head sees one relationship pattern. Twelve heads see twelve. Multi-head attention splits d_model across n_head parallel projections, runs attention in each head, then concatenates results back. ANDREA-120M splits 768 across 12 heads of 64 each. This activity walks through the split, the parallel computation, the concat, & why head specialization emerges from training pressure alone.

machine learning

Grow a Language Model: The Transformer Block

A transformer block stacks two sublayers: multi-head attention & a feed-forward MLP, each wrapped in a residual connection & layer normalization. ANDREA-120M stacks twelve such blocks. This activity walks through the pre-norm structure, why residuals matter, the MLP's 4× expansion ratio, the GELU activation, & how a 12-layer stack composes into a working language model.

machine learning

Grow a Language Model: Loss, Perplexity & The SMMA Curve

Cross-entropy loss measures how surprised the model gets when it sees the correct next token. Perplexity = exp(loss) measures the same surprise on a friendlier scale. SMMA (smoothed moving average) loss tracks training progress in real time. ANDREA-12M reached SMMA 2.0; ANDREA-120M descended to 3.43 at v1 minimum then polished to coherent multi-paragraph chat. This activity walks through the math, the random-chance baseline, & how to read a real loss curve.

machine learning

Grow a Language Model: Backprop In Custom CUDA

ANDREA's training engine `microgpt_cuda.cu` ships hand-written forward & backward CUDA kernels for every operation: attention, MLP, layer norm, embeddings. No PyTorch, no JAX, no autograd library. The chain rule lives in C-style code that gets compiled with nvcc & linked into a single binary. This activity walks through forward kernels, backward kernels, gradient accumulation buffers, & why writing your own framework gave ANDREA full control of memory & numerical precision on a single RTX 4090.

machine learning

Grow a Language Model: AdamW & Decoupled Weight Decay

Activity 10 of 25. AdamW separates weight decay from a gradient. Vanilla Adam collapsed ANDREA-120M v1 into repetition loops; weight_decay=0.01 in v2 produced stable training & factual recall. Walk through one optimizer step: first moment m, second moment v, bias correction, decoupled decay, parameter update.

machine learning

Grow a Language Model: LR Warmup & Cosine Decay

Activity 11 of 25. Learning rate schedule decides whether ANDREA reaches coherence or stalls in a bad basin. v1 used instant peak learning rate on fresh weights & collapsed; v2 ramps linearly over 2000 steps then decays cosine to zero. ANDREA-12M used a warm restart at step 25K. Walk through every regime.

machine learning

Grow a Language Model: Gradient Clipping

Activity 12 of 25. Gradient clipping caps the L2 norm of every gradient at max_norm=1.0 before AdamW touches a parameter. ANDREA-120M v1 had no clipping; source switches every 7-42 steps spawned gradient spikes that drove the model into degenerate attractors. Three CUDA kernels compute global L2 norm in parallel.

machine learning

Grow a Language Model: FP32, FP16, FP8 Precision

Activity 13 of 25. Numerical precision sets the memory footprint, training speed, & numerical stability of a language model. FP32 baselines safety; FP16 doubles throughput on tensor cores; FP8 E4M3 doubles again with NaN risk. ANDREA-120M uses FP16 cuBLAS at 6 steps/min on RTX 4090.

machine learning

Grow a Language Model: Multi-Armed Bandits & UCB1

ANDREA picks training data using a multi-armed bandit. Each data source becomes an arm; each step pulls one arm & gets reward. UCB1 (Auer, Cesa-Bianchi & Fischer, 2002) balances exploration against exploitation through a confidence-bound formula. This activity walks UCB1 from first principles, computes scores by hand, & explains why ANDREA tunes C=0.5 instead of a textbook 1.4.

machine learning

Grow a Language Model: Phase-Based Dice Control

Vanilla UCB1 picks one arm at a time. ANDREA wraps UCB1 inside phase-based dice control: variable phase lengths (7, 14, 21, 28, 42 steps), random arms first, UCB fills remaining focus slots, & 25-33% of phases run fully random. The 42-step ceiling caps damage from bad focus picks. This activity walks the dice math, the phase ladder, & why this structure prevents lock-in.

machine learning

Grow a Language Model: Reward Attribution & EMA

UCB1 needs a reward signal. Where does ANDREA get one? Each CUDA step reports (source, doc_index, loss). Proxy maintains a per-source exponential moving average of loss. Reward = max(0, EMA(t-1) - loss(t)) * 1000. The 1000x scaling matters: without it, per-step loss improvements (~0.001) get drowned by UCB exploration bonus (~0.5). This activity walks the reward pipeline & shows the math behind the scaling factor.

machine learning

Grow a Language Model: Source Floors & Epoch Penalty

UCB1 ranks arms; phase-based dice control assembles focus sets. But raw bandit output can starve priority sources or memorize tiny ones. ANDREA layers two protective rules on top: source floors guarantee minimum sampling weight per source, & epoch penalties downweight sources that have been pulled more than once per document (1/(1+epochs)). Lifetime pulls persist across restarts. This activity walks both rules with worked examples.

machine learning

Grow a Language Model: Coverage Bonus

Doc-level coverage tracking & a 1.3x freshness bonus push a bandit toward unexplored data. CUDA reports which document index sampled each step; a proxy tracks unique doc indices per source. Sources below 50% coverage earn a multiplier scaling toward 1.3x. This activity walks the math, contrasts coverage bonus against epoch penalty, & shows how doc-level freshness keeps a 500K-document gutenberg corpus from collapsing into a 2K-document subset during 200K training steps.

machine learning

Grow a Language Model: Curriculum Warmup

Restrict a firehose to 7 chat & prose sources for the first 20K steps, then activate the full mix. v1 (16 sources from step 0) collapsed into repetition; v2 added warmup & survived; v3 polish (after step 112K) sets curriculum_warmup_steps=0 because the model already learned coherent language. This activity explains why early diversity hurts a freshly initialized 120M model & how a two-phase curriculum balances coherence with breadth.

machine learning

Grow a Language Model: Filter by Shape, Not Characters

v2 trained on contaminated data: harness session JSONL leaked agent system prompts into the user-turn slot. The model learned that users speak in multi-section markdown & reproduced agent ornaments. v2.5 patch: `has_system_prompt_shape()` detects leaked prompts by SHAPE (header counts, length, fingerprint phrases), not by character matching. Three signals combine into a drop decision. Drop rates after re-filtering hermes3 sources: 87.7%, 90.4%, 93.0%. This activity walks the detector & the lesson it teaches: ornaments evolve, structure stays.

machine learning

Grow a Language Model: Coherence-Gated Early Stopping

v1 produced `region region region` for 10+ days because eval_chat_quality() wired only to a legacy multi-phase runner & never ran during the firehose curriculum. v2 ships a coherence gate: every sample scores on four metrics (bigram diversity 0-35, trigram diversity 0-35, English word presence 0-20, character diversity 0-10). Auto-halt after 5 consecutive samples score below 30. Back-tested on v1, the gate would have triggered at step 132K, saving 3.8 days of compute. This activity walks the four scorers, the consecutive-counter design, & why composite metrics catch failures a single signal misses.

machine learning

Grow a Language Model: Checkpoints, Resume, Signals

ANDREA writes a checkpoint every 100 steps. Each one packs a 4-byte step header, a 4-byte parameter count, then three float32 arrays back-to-back: weights, Adam first moment m, Adam second moment v. SIGTERM triggers an immediate checkpoint then exit; SIGUSR1 triggers an on-demand checkpoint without stopping. .loss.json never gets archived (cumulative training history per the never-archive rule). Walk the byte layout, sketch a resume, & lock in why one rule (do not archive .loss.json) outranks every other archival decision.

machine learning

Grow a Language Model: Sample Audit & External Grading

Loss is a number. Reading samples is how we know what that number means. Every 100 steps, ANDREA generates 420 tokens of free-form output, & loss alongside each sample lands in five tiers: 0.30 = factual recall (ppl 1, memorized), 0.74 = textbook one-liner (ppl 2), 1.05 = constraint following (ppl 3), 1.94 = knowledgeable but imprecise (ppl 7), 0.13 = anomalous memorization (zombie arm signal). Six unrelated knowledge domains in 700 steps prove the bandit working. External chat-quality grading rated samples 9.5/10. Walk a real audit, learn the loss tiers, & catch a zombie arm in the act.

machine learning

Grow a Language Model: From microGPT to ANDREA-120M

ANDREA-120M did not converge on its first try. v1 collapsed into 'region region region' & ate 165K of 200K planned steps. Five compounding failures. v2 added five fixes & ran two weeks before data contamination surfaced at step 15K. v2.5 patched the filter. v3 ran clean for 112K steps until a zombie bandit arm reproduced repo-docs verbatim, triggering the polish pivot at step 112,619. Within hours, samples rated 9.5/10. Walk the engineering record from collapse to coherence; learn what each fix solved & what the next failure taught.

math

Geometry in the Real World: From Architecture to Machine Learning

How geometry shapes architecture, computer graphics, machine learning, navigation, and engineering.

math

Geometry of Capital

Radar charts, network graphs & flow diagrams: geometric tools for visualizing capital portfolios, measuring balance across eight dimensions & mapping flows between forms.

math

Geometry of Kanban

Queuing theory, flow diagrams, and cumulative flow curves: the geometric thinking behind WIP limits, Little's Law, and bottleneck analysis in cross-functional work systems.

music

Geometry of Woodwinds

Pipe acoustics and why length sets pitch, hole placement as a geometric sequence, why a clarinet overblows a twelfth while a flute overblows an octave, and the circle of fifths as a literal circle: the geometry inside every wind instrument.

music

Recorder to the Woodwind Family

Why the recorder is the cheapest, most durable on-ramp into a whole instrument program: what transfers to the flute, clarinet, oboe, and saxophone, what is new, teaching beginning recorder, and the parallel on-ramps of percussion, piano, and guitar.

nursing

Nursing: Patient Care Fundamentals

Science & art of nursing: from vital signs to patient advocacy, bedrock of healthcare careers.

physics

Geometry of Nuclear Physics

Reactor core lattice geometry, neutron flux distribution, shielding calculations, and the inverse square law.

physics

Geometry of Orbital Mechanics

Elliptical orbits, Hohmann transfers, orbital planes, inclination changes, and Lagrange points: the geometry of spaceflight.

physics

Nuclear Engineering 401 -- Reactor Design Capstone

A full semester capstone project: design a safe nuclear reactor from first principles. Choose reactor type, fuel, coolant, triple-redundant safety systems, human oversight, siting, and licensing strategy: and defend every decision like a professional engineer.

physics

Nuclear Physics 101: The Atomic Nucleus

Community college-level nuclear physics. Strong force, shell model, radioactive decay, binding energy, fission, fusion, and E=mc²: with real calculations.

physics

Nuclear Physics 201: Reactor Principles

Community college-level nuclear engineering: neutron physics, criticality, four-factor formula, fuel cycles, control rod physics, thermal hydraulics, reactor types & reactor accidents that rewrote safety culture.

physics

Nuclear Physics 301 -- Safety Systems and Defense in Depth

A full community college safety engineering module: defense in depth, triple redundancy, ECCS, containment, reactivity control, electrical systems, I&C, human factors, ALARA, accident lessons, PRA, and waste management.

plumbing

Geometry of Plumbing

Pipe grade calculations, fitting angles, offset geometry, drainage slope, and the physics of water flow in geometric systems.

plumbing

Plumbing: Water In, Waste Out

Understand water supply and drainage systems: the invisible infrastructure that makes modern life possible.

robotics

Geometry of Robotics

Forward and inverse kinematics, workspace envelopes, degrees of freedom, and path planning: the geometry that makes robots move.

robotics

Robotics: Machines That Move

Sensors, actuators, and control loops: how robots perceive, decide, and act, from factory arms to Mars rovers.

unhamming

Unhamming: What Hamming Gave & What He Missed

Richard Hamming built a dragon of knowledge from Bell Labs & the cold war era. Unhamming grows from his back: same fire, different flight. No spy versus spy. No predator & prey. A mapping of what Hamming got right, what his era blinded him to, & what a permacomputer extends.

unhamming

The Missing Chapter: Algorithmic Complexity

Hamming covered digital filters, coding theory, simulation, & n-dimensional geometry. He never named Big O. This lesson fills the gap: O(1) through O(N²) with concrete programming examples, the cost of silence, & why MOAD-0001 grew in Hamming's era unnoticed.

unhamming

MOAD-0001: The Sedimentary Defect

A list used where a set belongs. CWE-407: O(N²) membership tests in 1,000+ sites across 60+ software ecosystems. Hamming's era seeded it; our era inherits it. This lesson teaches pattern recognition, code archaeology, & the one-line fix that makes infrastructure 1,000× faster.

unhamming

Undefect: Finding MOADs in the Wild

Knowing a MOAD exists differs from finding it. This lesson teaches the detection methodology: how to read code for sedimentary patterns, how to search ecosystems with grep & static analysis, & how to move from confirmed defect to upstream patch using a MOAD pipeline: scan, ticket, patch, test, disclose, PR, merge.

unhamming

Eight Forms of Capital: What Hamming Never Measured

Hamming taught 'You Get What You Measure' — one of his best insights. He measured financial capital & intellectual capital. Eight forms of capital exist. This lesson extends his measurement framework to all eight, then asks: what does your work grow & what does it drain?

unhamming

Permacomputer: Infrastructure That Outlasts Its Builders

Russell Ballestrini wrote a permacomputer manifesto: infrastructure grows without extracting rent, code outlasts authors, & free compute belongs to everyone. This lesson extends Hamming's systems engineering (Ch. 28) to the permacomputer design principles: capability-driven presentation, progressive enhancement, & cooperative infrastructure.

unhamming

You & Your Research: Open-Source Edition

Hamming's 'You and Your Research' assumed institutional backing: a Bell Labs, a university, a national lab. Open-source removes that assumption. This lesson extends Hamming's best advice — work on important problems, keep 10 important problems in mind, compound your knowledge — to independent researchers with nothing but commit access & a public repo.

unhamming

Without Spy versus Spy

Hamming's era ran inside a cold-war binary: us versus them, classified versus open, Bell Labs versus everyone else. Unhamming operates without that frame. This lesson examines what the binary costs in each capital queue & what happens when you grow a board that does not need the game.

unhamming

Permaculture & Code: Growing versus Extracting

Permaculture principle: observe & interact; catch & store energy; produce no waste. Applied to software: write code that regenerates the systems around it rather than extracting from them. This lesson maps permaculture design principles to permacomputer infrastructure & a MOAD factory model.

unhamming

Balance All Workstations: MOAD-0001 & MOAD-0005 Are Coupled

Hamming's systems engineering (Ch. 28): optimizing one component in isolation degrades the whole system. A MOAD factory model applies this at infrastructure scale: fix O(N²) at a high-throughput node & every downstream queue floods. MOAD-0001 & MOAD-0005 couple. Solve both before disclosing either.

unhamming

The Quadrivium: Truth, Freedom, Harmony, Love

Hamming ended 'You and Your Research' with the question every researcher must answer: what are your values, & what will you sacrifice for them? The permacomputer quadrivium gives a concrete answer: Truth (open source), Freedom (voluntary participation), Harmony (self-renewal), Love (wu wei). This lesson applies each to software & research practice.

unhamming

MOAD-0002: An Intertangled Defect

Subsystems coupled through shared mutable global state, mutated synchronously in a single event loop or god-object. A structural coupling, not a timing problem. This lesson teaches why adding a mutex makes it worse, & how phase-snapshot decouples subsystems at the architectural level.

unhamming

MOAD-0003: A Leaked Context

ThreadLocal holds request-scoped identity. Thread returns to pool. ThreadLocal.remove() not called. Next request on the same thread reads the previous request's context. Not a crash; a silent security boundary violation. This lesson teaches the scope mismatch & the lifetime-correct primitives that replaced it.

unhamming

MOAD-0004: A Logged Secret

Web servers log HTTP headers verbatim. Authorization, Cookie, & X-API-Key get written to disk in cleartext. Not in memory; on disk, in log shippers, in aggregators, retained for 30-90 days. This lesson teaches the exposure window calculation & the credential denylist fix at the serialization layer.

unhamming

MOAD-0005: A Thundering Herd

cache.get(key) + null check + expensiveCompute(key) + cache.put(key): four non-atomic steps. N threads all miss, all compute simultaneously. The idempotency trap: 'it is OK if two threads write the same result' gets correctness right but ignores cost. This lesson teaches the atomic fix & its coupling to MOAD-0001.

unhamming

MOAD-0006: A Glass Safe

Credentials stored in recoverable form: plaintext, reversible encryption, base64. A single database dump exposes all credentials simultaneously. Unlike MOAD-0004 (accidental log write), a Glass Safe requires credential recovery by architectural design. This lesson teaches why strong encryption does not fix it, & what one-way hash replaces.

unhamming

MOAD-0007: A Flatland Defect

Spatial data queried with O(N) linear scan. Geometry structure discarded. BVH, octree, & k-d tree never built. Named after Edwin Abbott's Flatland (1884): the structure exists; it gets discarded. Three.js Raycaster at 60Hz with 10,000 objects: 600,000 geometry checks per second from a 3-line fix. This lesson teaches the spatial index & why a hash set cannot replace it.

unhamming

MOAD-0009: A Metered Heart

A scheduled job fires on a timer regardless of whether work needs doing. Two forms: a state-repair job that fixes broken state after a delay (users see glitches during the window), & a wasteful report that recomputes from scratch on a fixed interval. This lesson teaches event-driven design as the replacement & the MOAD amplification pattern.

unhamming

Why MOAD Festers: The History of Known Defects

Five defects — O(N²) membership, global state coupling, ThreadLocal identity leak, credential logging, thundering herd — were knowable decades before systematic detection in 2026. This lesson traces the intellectual history, identifies the five structural conditions that allowed each to persist, & builds a diagnosis vocabulary for recognizing unfixed knowable defects.

welding

Geometry of Welding

Joint angles, bevel geometry, weld profiles, distortion control, and fit-up tolerances: geometric precision in metal joining.

welding

Welding Fundamentals: Joining Metal

Learn science & craft behind welding: from arc physics to joint design, foundation of every trade career in metal fabrication.

Unlimited access — no payment required

Complete 2 lessons, write a paragraph about your enterprise & earn solo-tier access through entrepreneur contribution track.

Entrepreneur track →