Further reading
References, datasets, models, and software used across the four courses.
Further reading establishes the research, equations, standards, and APIs behind each lesson. Its prose, figures, tutorial code, and datasets are not republished here; Latent's explanations, diagrams, exercises, implementations, and synthetic fixtures are course-authored.
Models and open-source software
SmolLM2-135M-Instructupstream model
Apache-2.0 ↗SmolLM2-135M-Instruct-ONNXONNX conversionupstream Apache-2.0 model ↗Transformers.js3.8.1Apache-2.0 ↗React + React DOM19.2.6MIT ↗Dexie4.4.4Apache-2.0 ↗CodeMirror6.0.2MIT ↗CodeMirror Python language6.2.1MIT ↗Pyodide314.0.2MPL-2.0 ↗NumPy2.4.3BSD-3-Clause ↗esbuild-wasm0.28.1MIT ↗QuickJS Emscripten0.32.0MIT ↗Mock Service Worker2.15.0MIT ↗Further reading by lesson
Linear Algebra Basics5 lessons
Small arrays · Course-authored synthetic examples · Not separately licensed · 6 fixed numerical examplesTwo- and three-coordinate vectors · Course-authored synthetic examples · Not separately licensed · 8 fixed numerical examplesVector pairs · Course-authored synthetic examples · Not separately licensed · 9 fixed vector pairsSmall weight matrices · Course-authored synthetic examples · Not separately licensed · 8 fixed matrix and vector combinationsSmall vector batches · Course-authored synthetic examples · Not separately licensed · 7 fixed batch and bias examplesMachine Learning Basics5 lessons
Small Labeled Table · Course-authored synthetic examples · Not separately licensed · 8 rows · 2 features · 1 targetSmall Regression Set · Course-authored synthetic examples · Not separately licensed · 5 examples · 2 numeric featuresScalar Regression Set · Course-authored synthetic examples · Not separately licensed · 4 examples · 1 weight · 1 biasBinary Score Set · Course-authored synthetic examples · Not separately licensed · 6 logits · binary targetsSmall Forward-Pass Set · Course-authored synthetic examples · Not separately licensed · 4 inputs · 2 hidden units · 1 outputProbability · calculus · regularization · evaluation- Mathematics for Machine LearningMarc Peter Deisenroth · A. Aldo Faisal · Cheng Soon Ong · 2020↗
- Elements of Information TheoryThomas M. Cover · Joy A. Thomas · 2006↗
- 18.02SC Multivariable CalculusMIT OpenCourseWare · 2010↗
- Convex OptimizationStephen Boyd · Lieven Vandenberghe · 2004↗
- Deep Learning: Regularization for Deep LearningIan Goodfellow · Yoshua Bengio · Aaron Courville · 2016↗
- Dropout: A Simple Way to Prevent Neural Networks from OverfittingNitish Srivastava et al. · 2014↗
- Decoupled Weight Decay RegularizationIlya Loshchilov · Frank Hutter · 2019↗
- Model evaluationscikit-learn developers · Current↗
- Probability calibrationscikit-learn developers · Current↗
- NIST/SEMATECH e-Handbook of Statistical MethodsNIST · SEMATECH · 2012↗
- An Introduction to the BootstrapBradley Efron · Robert J. Tibshirani · 1993↗
Harness Engineering8 lessons
Recorded agent responses · Course-authored synthetic traces · Not separately licensed · 8 response and tool-result fixturesTool-call fixtures · Course-authored synthetic examples · Not separately licensed · 10 argument and result-page fixturesContext-budget fixtures · Course-authored synthetic traces · Not separately licensed · 9 instruction, file, test, and tool-output itemsPermission-policy fixtures · Course-authored synthetic examples · Not separately licensed · 12 paths, actions, and overlapping policy rules- Agent approvals & securityOpenAI · Current↗
- SandboxingOpenAI · Current↗
- Understanding prompt injectionsOpenAI · Current↗
- LLM01:2025 Prompt InjectionOWASP GenAI Security Project · 2025↗
- Custom instructions with AGENTS.mdOpenAI · Current↗
- CWE-22: Improper Limitation of a Pathname to a Restricted DirectoryMITRE · Current↗
- CWE-367: Time-of-check Time-of-use Race ConditionMITRE · Current↗
Run-event fixtures · Course-authored synthetic traces · Not separately licensed · 10 completion, checkpoint, failure, and finish eventsAgent-evaluation fixtures · Course-authored synthetic outcomes · Not separately licensed · 8 outcomes, requirements, and trial sets- SWE-bench: Can Language Models Resolve Real-World GitHub Issues?Carlos E. Jimenez et al. · 2024↗
- Scoring metricsUK AI Security Institute · Current↗
- Evaluating Large Language Models Trained on CodeMark Chen et al. · 2021↗
- tau-bench: A Benchmark for Tool-Agent-User Interaction in Real-World DomainsShunyu Yao et al. · 2024↗
Task-graph fixtures · Course-authored synthetic examples · Not separately licensed · 9 acyclic, cyclic, and incomplete task graphsIntegrated harness traces · Course-authored synthetic examples · Not separately licensed · 11 complete, denied, approval-gated, and malformed runsBuild an LLM System in Your Browser14 lessons
Signal Notes · Course-authored synthetic corpus · Not separately licensed · 1,610 characters · fixed repeatable sequenceRoles and Actions · Course-authored synthetic corpus · Not separately licensed · 20 sentences · fixed example-by-example modulo splitMorphology Set · Course-authored synthetic corpus · Not separately licensed · 6 lines · 24 words · fixedDate Alignment · Course-authored synthetic task · Not separately licensed · 3 semantic roles · 3 fixed alignment cases · 2,000 epochsCausal Sequence Set · Course-authored synthetic examples · Not separately licensed · 1 fixed six-token sequenceOpaque Review Labels · Course-authored synthetic evaluation set · Not separately licensed · 4 demonstrations · 2 held-out casesInference Trace · Course-authored synthetic requests · Not separately licensed · 6 requests · fixed prompt and output lengthsServing Workload · Course-authored synthetic arrivals · Not separately licensed · 9 requests · mixed prompt and output lengthsToken Event Trace · Course-authored synthetic stream · Not separately licensed · 14 frames · adversarial chunk boundariesFailure Trace · Course-authored synthetic scenarios · Not separately licensed · 4 failure modes · fixed seedsConversation Event Log · Course-authored synthetic actions · Not separately licensed · 18 reducer actions · 3 generation attemptsRender Trace · Course-authored synthetic stream · Not separately licensed · 60 deltas · 4 timing profilesBranching Conversation · Course-authored synthetic scenario · Not separately licensed · 3 action flows · 29 budgets (14–42)Product Contract Audit · Course-authored synthetic checklist · Not separately licensed · 11 executable pure checks · 5 specifications · 3 manual verification groups