A comprehensive, open academic course on LLM prompt engineering covering the full 2026 model ecosystem — frontier proprietary (Claude, GPT, Gemini, Grok), Chinese frontier (DeepSeek, Kimi, MiniMax, Qwen, GLM), open-weights (Llama, Qwen, Gemma, Mistral, DeepSeek, Phi), and local deployment (Ollama, llama.cpp, MLX, vLLM).
Vendor-agnostic by default. Cross-API translation tables for every technique. Bilingual ES + EN. Versioned content with changelog. Comparable in scope to Anthropic Academy, Google ML Crash Course, DeepLearning.AI Generative AI courses — without vendor lock-in.
🔗 Live demo: gs-run.github.io/prompt-engineering-course
📋 Current version: 2.3.16 (see CHANGELOG)
The course is organised into 14 academic blocks + an orientation block + workshop + capstone. Each block is a standalone HTML page (~7-67 KB) loaded on demand, opening with front-matter (icon, level, time estimate, learning objectives) and closing with bibliographic references + a 5-question knowledge check.
| # | Block | Level | Time |
|---|---|---|---|
| 0 | Orientation (setup, prerequisites) | beginner | 20 min |
| I | Foundations (LLM mechanics, ecosystem) | beginner | 90 min |
| II | Prompt Engineering Core | beginner | 180 min |
| III | Advanced Reasoning + Tools | intermediate | 150 min |
| IV | Production Engineering | intermediate | 200 min |
| V | Agents + Multi-agent | intermediate | 180 min |
| VI | Infrastructure + RAG deep-dive | intermediate | 150 min |
| VII | Local Models + Quantization + Privacy | advanced | 200 min |
| VIII | Cross-Model Patterns | advanced | 60 min |
| IX | Benchmarks + Evaluation | intermediate | 90 min |
| X | Safety, Ethics, Regulation | intermediate | 90 min |
| XI | Applied Industry + Roles | intermediate | 200 min |
| XII | Future + Research Frontier | reading | 60 min |
| XIII | Practical Workshop (Bando B) | all | 180 min |
| XIV | Capstone Projects (4 end-to-end) | capstone | ~10h × 4 |
Total study time: ~30 hours across 4-6 weeks at 2-3 hour sessions.
- Token Counter — real-time token estimation.
- Cost Calculator — 18 frontier + OSS models grouped by region (US / China / EU), with caching scenarios.
- Prompt Diff Comparator — A/B two prompts with 5-criteria scoring.
- 🪄 SKILL.md Builder — fill in name, description, allowed-tools, context and body; copy the generated SKILL.md ready to drop into
~/.claude/skills/<name>/. - 🎯 Agent Goal Composer — compose a structured agent prompt with role, goal, constraints, success criteria and output format (Anthropic-style XML tags).
- 🩺 Prompt Linter — paste a system prompt and get 7 checks + score (length, role, examples, format, XML structure, CoT, user-data leak) based on Anthropic's prompt-engineering guide.
- Prompt Simulator — toggle 6 techniques and see simulated response quality.
- 5 Coding Exercises with real-time scoring.
- 92 quizzes total — 22 inline formative + 70 end-of-block knowledge checks, distributed 23/23/23/23 across A/B/C/D (rebalanced in 2.2.3).
- Glossary — 89 technical terms ES + EN with rich schema (short tooltip + long-form definition + examples + related-term chips). 73 entries carry a hand-tuned inline SVG diagram (RAG flow, attention weights, agent loop, MoE routing, transformer stack, KV cache, LoRA decomp, fine-tuning vs RAG, sampling distributions, training timeline, prompt injection flow, frontier scatter, MCP topology, and 60+ more). In-text auto-link wraps every detected term in a clickable link that opens the glossary in a new tab.
git clone https://github.com/GS-RUN/prompt-engineering-course.git
cd prompt-engineering-course
npx serve .
# open http://localhost:3000Or just open index.html directly in a browser — no build step, no dependencies.
prompt-engineering-course/
├── index.html # Landing (35 KB) — hero + 17 block cards + course philosophy
├── modules/
│ ├── 00-orientation.html # Setup, prerequisites, environment
│ ├── 01-foundations.html # LLM mechanics + 2026 ecosystem
│ ├── 02-prompt-core.html # Anatomy, roles, clarity, examples, XML, CoT, output
│ ├── 03-advanced-reasoning.html # Thinking, tool use, multimodal
│ ├── 04-production.html # Structured outputs, caching, evals, security, costs
│ ├── 05-agents.html # Skills, sub-agents, MCP, multi-agent patterns
│ ├── 06-infrastructure.html # Chunking, vector DBs, deployment, monitoring
│ ├── 07-local-quant.html # GPTQ/AWQ/GGUF, fine-tuning, privacy
│ ├── 08-cross-model.html # Vendor-agnostic patterns, fallback, routing
│ ├── 09-benchmarks.html # MMLU/HumanEval/SWE-bench/ARC-AGI + eval design
│ ├── 10-safety.html # Hallucination, red teaming, EU AI Act
│ ├── 11-industry.html # Code review, support, extraction, by sector + role
│ ├── 12-future.html # Research frontiers, expected models, AI + science
│ ├── 13-workshop.html # Bando B: project, linter, library, evolution
│ ├── 14-capstone.html # 4 end-to-end project specs with rubrics
│ ├── glossary.html # 89 terms ES + EN, rich schema, anchored, in-text auto-linking
│ └── tools.html # 6 tools: tokens, cost, prompt diff, SKILL.md, agent goal, linter
├── css/
│ ├── style.css # Carbón + Ámbar palette, dark + light themes
│ └── bando-b.css # Workshop-specific styling
├── js/
│ ├── shared/
│ │ ├── manifest.js # Single source of truth (block list, objectives, sections)
│ │ ├── sidebar.js # Generates sidebar dynamically per page
│ │ ├── landing.js # Renders block grid on home page
│ │ ├── references.js # Per-block bibliography
│ │ └── glossary-data.js # 89 term definitions with rich schema (short + long + examples + related)
│ ├── app.js # Theme controller, defensive widget instantiation
│ ├── i18n.js # ES + EN toggle (CSS-driven via lang-blocks)
│ ├── quiz.js # 92 multiple-choice quizzes with schema-versioned localStorage migration
│ ├── exercises.js # 5 interactive scoring exercises
│ ├── simulator.js # Technique-toggle prompt simulator
│ ├── diagrams.js # Animated Canvas diagrams (legacy block visuals)
│ ├── glossary-diagrams.js # 73 hand-tuned inline-SVG diagrams for the glossary (RAG flow, attention, MoE, KV cache, etc.)
│ ├── glossary.js # In-text auto-link wrapping → links to glossary.html in a new tab + hover tooltip
│ ├── glossary-page.js # Renderer for the dedicated glossary page (anchored entries, smooth scroll, search)
│ ├── token-tools.js # Token counter + cost calculator
│ ├── prompt-diff.js # A/B prompt comparator
│ ├── tools-extra.js # SKILL.md Builder + Agent Goal Composer + Prompt Linter
│ ├── linter.js # Prompt linter (Bando B workshop, separate from the v2.3.16 Prompt Linter tool)
│ ├── library.js # Prompt library (Bando B)
│ ├── evolution.js # Prompt evolution tracker (Bando B)
│ └── project.js # User project panel (Bando B)
├── scripts/
│ ├── split_modules.js # One-shot: extract sections from index.legacy.html into modules/
│ ├── add_knowledge_checks.js # One-shot: insert kc-NN anchors into all module pages
│ ├── wrap_lang_blocks.js # One-shot: wrap section bodies in lang-block divs
│ └── rebalance_quiz_answers.js # One-shot: rebalance correct-answer positions across all quizzes (used in v2.2.3)
├── CHANGELOG.md
├── README.md
└── index.legacy.html # v1 monolithic backup (will be removed in a future major)
- Multi-page split: initial landing 35 KB; each block 7-67 KB. Browser caches CSS/JS across pages.
- Single source of truth:
js/shared/manifest.jsdeclares the block structure. The sidebar, landing, and split scripts all consume it. - Bilingual via CSS: every section body is wrapped in
<div class="lang-block" data-lang="es|en">. The<html>element'sdata-i18n-langattribute drives visibility. No JS replacement of textContent. - Defensive JS instantiation: widget engines are only constructed if their class is loaded on the current page (each module imports only the JS it needs).
- Versioned content: every claim is dated 2026-01 to 2026-05. The "model + hardware version policy" in Block XIV instructs readers to substitute upward (e.g. Opus 4.7 → Opus 5+ when shipped).
Click ES / EN in the sidebar to switch. Coverage:
- Every section body has parallel ES + EN content. Sections without an EN translation fall back to ES with a banner.
- Sidebar, landing cards, glossary, knowledge checks, references — all bilingual.
- Language preference saved to
localStorageand applied via<html data-i18n-lang="es|en">. - Convention: prompt examples appear in the active language. The course flags honestly that production prompts often run 5-15% better in English on small open-weights models.
- Carbón + Ámbar palette (warm dark default, light theme available).
- CSS-only animated background (gradient orbs, subtle grid).
- Glassmorphism cards with backdrop blur.
- Responsive: sidebar collapses to a horizontal bar on mobile.
- Difficulty badges (beginner / intermediate / advanced / capstone / reference / reading).
- Animated progress bar at the top.
The course is researched and compiled from official documentation. Each block also has its own per-block bibliography in js/shared/references.js with direct links to papers, RFCs and official docs.
Primary sources (provider documentation):
- Anthropic — Prompt Engineering Guide
- Anthropic — Claude Code Docs
- OpenAI — Prompt Engineering
- Google — Gemini Prompt Best Practices
- DeepSeek — API Docs
- Moonshot — Kimi Open Platform
- Alibaba — Qwen / DashScope Docs
- Zhipu — BigModel / GLM Docs
- Mistral — La Plateforme Docs
- Meta — Llama Cookbook
- Microsoft — Phi Cookbook
- Cohere — Documentation
- OpenCode Docs
- DAIR.AI — Prompt Engineering Guide
Foundational papers (linked in per-block references):
- Vaswani et al. — Attention Is All You Need (2017)
- Brown et al. — Language Models are Few-Shot Learners (GPT-3, 2020)
- Wei et al. — Chain-of-Thought Prompting (2022)
- Yao et al. — ReAct: Synergizing Reasoning and Acting (2022)
- Bai et al. — Constitutional AI (Anthropic, 2022)
- Frantar et al. — GPTQ (2022); Lin et al. — AWQ (2023); Dettmers et al. — QLoRA (2023)
- Hendrycks et al. — MMLU (2020); Chen et al. — HumanEval (2021); Rein et al. — GPQA (2023)
- Greshake et al. — Indirect Prompt Injection (2023)
Pull requests welcome. The course is versioned (semver-ish): patch for fixes/typos, minor for new content, major for architectural changes.
Architecture contracts:
- Block structure → edit
js/shared/manifest.js. Re-runnode scripts/split_modules.jsif changing legacy section ranges. - Knowledge checks → add to
js/quiz.jswithsection: "kc-NN". - Glossary terms → add to
js/shared/glossary-data.js. - Per-block references → edit
js/shared/references.js. - New module page → use the same shell as existing pages (
<head>includes,<body data-page-block="NN">,<nav id="sidebar">populated bysidebar.js,<main id="content">, references section, knowledge-check anchor).
When to bump version:
- 2.2.1 → 2.2.2: typos, refactors, link fixes.
- 2.2.x → 2.3.0: new section, new module, content rewrite.
- 2.x.y → 3.0.0: architectural change (manifest schema, breaking URL changes).
Always update CHANGELOG.md with the change. Version lives in js/shared/manifest.js and is shown in the sidebar foot + landing footer.
MIT — use it, modify it, share it.
© 2026 Alonso J. Núñez · GS·RUN · Source code