LLM Engineering and Agentic AI Expert Program

Self-study repository covering, in depth and with real-world practice, the full syllabus of an LLM Engineering expert program (10 ECTS): LLM fundamentals, prompt engineering, RAG systems, agents, LLMOps, and a production-level capstone project. Includes preparation for the AWS Certified AI Practitioner (AIF-C01) and NVIDIA Certified Associate: Generative AI LLMs (NCA-GENL) certifications.

Current Program Stack#

Stack snapshot reviewed on August 21, 2026. The executable defaults are gpt-5.6-luna and claude-haiku-4-5; the routing module compares GPT-5.6 Luna, Terra, and Sol. The panorama includes Claude Sonnet/Opus 5, Gemini 3.x, Llama 4, Mistral Small 4, and the current Qwen family. GPT-4o, GPT-3, and Llama 3 appear only when needed to explain history, papers, or tokenizers—never as defaults for a new integration.

New code uses current OpenAI Responses, structured outputs, and function calling; Anthropic Messages/tool use; LangGraph 1.2; MCP Python SDK 2.x; and RAGAS 0.4.3. All models are configurable via .env: updating the catalog requires neither code changes nor invalidating evaluation datasets.

Repo Structure#

Folder	Content	ECTS
modulo-01-fundamentos-llm/	Transformer architecture, tokenization, generation parameters, model landscape, Amazon Bedrock, and OpenAI/Anthropic APIs	0.5
modulo-02-prompt-engineering/	Zero/few-shot, chain-of-thought, function calling, structured outputs (Pydantic + Instructor), prompt evaluation and versioning	1.5
modulo-03-rag/	Production RAG: chunking, vector DBs, reranking, advanced RAG (HyDE, self-RAG, CRAG), RAGAS evaluation, full-stack app	2
modulo-04-agentes/	Agent patterns (ReAct, reflection), LangGraph, MCP, tool use, multi-agent, memory, Bedrock Agents, alignment	2
modulo-05-llmops/	Observability, continuous evaluation, cost optimization, Docker/K8s, vLLM/Ollama, AWS, Responsible AI, security	2
modulo-06-capstone/	Final project: RAG + autonomous agent deployed on cloud with LLMOps and controlled costs	2
certificaciones/	Study guides, flashcards, and mock exams for AIF-C01 and NCA-GENL	—
setup/	Development environment, dependencies, and API keys	—
recursos/	Bibliography, papers, courses, and annotated links	—

How to Use This Repo#

1. Set up the environment by following setup/README.md (Python 3.12 + uv, API keys in .env).
2. Follow the modules in order. Each module includes:
   • README.md — module map and learning objectives
   • teoria/ — in-depth notes, one file per topic
   • labs/ — executable code, one lab per key concept
   • ejercicios.md — proposed exercises with acceptance criteria and public tests

Reference solutions are not part of the public repository. Grading relies on tests, rubrics, and reproducible evidence so that completing an exercise demonstrates competence. CI runs node scripts/validate-public-content.mjs and blocks any leakage via path or reserved marker. 3. Work on certifications in parallel: AIF-C01 is covered in modules 1–5 and NCA-GENL in modules 2–6. Mock-exam questions are in certificaciones/; the reasoned grading key is integrated from the website's private source. 4. Finish with the capstone in module 6, which integrates everything above.

The suggested schedule and full syllabus details are in PLAN_DE_ESTUDIOS.md.

Web Version and Progress#

llmengineerclub.com transforms this content into an Astro bilingual experience with a guided path, search, local tools, mock exams, flashcards, a skills assessment, and portfolio. No account required: progress lives in the browser. Page authorship is synchronized from merged PRs and links to the public profile of whoever created, improved, or reviewed the content.

The Club Quiz is a separate, voluntary experience: it offers a local practice run with retired questions and one official attempt per pseudonymous passkey. Scoring takes both correct answers and server-measured response speed into account. It uses no cookies and the result is private by default; only afterwards may a participant voluntarily link a social profile to enter the leaderboard. Every result has a verifiable Ed25519 proof.

This repository contains knowledge, problem statements, fictional data, and verifiable criteria. Editorial exercise solutions are kept outside both the public repository and the web artifact. Mock-exam keys do not live here either: the website combines them from its private source and only reveals the review after an exam is submitted.

PRs may also propose objectives and blueprints for future questions. The exact material for a competitive edition is transformed and reviewed outside the public history; after retirement, it may return here as open practice. Read the Club Quiz contribution policy.

Local Validation#

node scripts/validate-public-content.mjs
node scripts/validate-course-contracts.mjs
node scripts/validate-current-stack.mjs
node scripts/validate-certification-currency.mjs
node scripts/validate-translations.mjs --complete
uv sync --locked --extra dev
uv run ruff check .
uv run python -m compileall -q modulo-* setup translations/en
uv run pytest -q

Stack validation rejects legacy models as defaults and requires a course catalog review every 120 days. Certification validation pins the published contracts for AIF-C01 revision 1.1 and NCA-GENL and rejects retired nomenclature. Bilingual validation preserves code blocks, link destinations, and the executable AST of each lab.

Consult CONTRIBUTING.md before opening a PR. Every merged contribution retains permanent attribution on the web.

Prerequisites#

• One year of full-stack programming experience (Python is the repo language).
• ML fundamentals: model, training, validation, overfitting, metrics.
• REST APIs: HTTP, JSON, API key authentication.
• VS Code or Jupyter and basic Git.
• Recommended: NumPy/pandas, SQL, some AWS experience, and prior hands-on use of ChatGPT or Claude in a technical context.

Professional roles covered by the syllabus#

LLM Engineer / AI Engineer · RAG Systems Engineer · AI Agents Developer · AI Product Engineer · AI Consultant/freelancer.