Curriculum — LLM Engineering and Agentic AI Expert (10 ECTS)

1 ECTS ≈ 25–30 hours of work. Estimated total: 250–300 hours. At a pace of 10 h/week, approximately 6–7 months; at 20 h/week, approximately 3 months.

Suggested Schedule (10 h/week pace)#

Weeks	Module	ECTS	Milestone
1–2	I. LLM Fundamentals and APIs	0.5	Working multi-provider client (OpenAI, Anthropic, Bedrock)
3–6	II. Advanced Prompt Engineering	1.5	Prompt evaluation pipeline with test dataset and A/B testing
7–12	III. RAG Systems and Evaluation	2	Full-stack RAG app with RAGAS ≥ 0.75 on faithfulness
13–18	IV. AI Agents and Orchestration	2	Multi-agent system with LangGraph + custom MCP server
19–24	V. LLMOps, Production, and Responsible AI	2	Module III/IV system deployed, monitored, and with measured costs
25–30	VI. Capstone	2	Complete cloud system + AIF-C01 and NCA-GENL practice exams

---

Module I — LLM Fundamentals and APIs (0.5 ECTS)#

Transformer architecture, tokenization, and generation parameters. Overview of current models and first exposure to Amazon Bedrock and the OpenAI and Anthropic APIs.

• Transformer architecture: encoder, decoder, and multi-head attention
• Tokenization: BPE, WordPiece, cross-model comparison
• Generation parameters: temperature, top_p, top_k, penalties
• Model landscape: GPT-5.6, Claude 5/4.5, Gemini 3.x, Llama 4, and Mistral Small 4
• AWS AI services: Bedrock, SageMaker JumpStart, and Amazon Q
• Amazon Bedrock: foundation models, inference, and API

Module II — Advanced Prompt Engineering (1.5 ECTS)#

From zero-shot to few-shot, chain-of-thought, function calling, and structured outputs with Pydantic and Instructor. Prompt evaluation using test datasets and A/B testing, prompt management and versioning, and bias and hallucination detection.

Note: the original brochure erroneously repeats the Module I bullets here; this repo expands on the actual syllabus described in the module paragraph.

• Core techniques: zero-shot, few-shot, chain-of-thought, self-consistency
• System prompts, roles, and context scoping
• Function calling / tool use in OpenAI and Anthropic
• Structured outputs: JSON mode, Pydantic, Instructor
• Prompt evaluation: test datasets, LLM-as-judge, A/B testing
• Prompt management and versioning (registries, templates, CI)
• Bias and hallucination detection and mitigation

Module III — RAG Systems and Evaluation (2 ECTS)#

Complete production-grade RAG architecture, from ingestion to generation. Rigorous evaluation with RAGAS and Bedrock Knowledge Bases as a managed service.

• RAG architecture: ingestion, embedding, indexing, and retrieval
• Vector databases: Pinecone, Weaviate, Qdrant, pgvector
• Chunking strategies: fixed, semantic, hierarchical, late
• Reranking: cross-encoders, Cohere Rerank, ColBERT
• Advanced RAG: HyDE, multi-query, self-RAG, CRAG
• Amazon Bedrock Knowledge Bases: managed RAG on AWS
• RAG evaluation: RAGAS, faithfulness, relevance, context recall
• Hallucination detection: metrics, tools, and mitigation
• Full-stack RAG app: FastAPI + Next.js + vector DB

Module IV — AI Agents and Orchestration (2 ECTS)#

Agents capable of planning, reasoning, and using tools. Multi-agent systems with LangGraph and MCP. Memory, flow control, and reliability evaluation.

• Agent patterns: ReAct, plan-execute, reflection, self-critique
• LangGraph: state graphs, cycles, conditionals, and checkpoints
• Model Context Protocol (MCP): architecture and server development
• Advanced tool use: tool design, error handling
• Multi-agent systems: coordination, supervisors, communication
• Agent memory: episodic, semantic, and working
• Amazon Bedrock Agents: action groups, integrated knowledge bases
• Agent evaluation: determinism, costs, and safety guardrails
• Alignment: RLHF, RLAIF, Constitutional AI, and fine-tuning techniques

Module V — LLMOps, Production, and Responsible AI (2 ECTS)#

Operation and deployment of LLM systems in production; observability, continuous evaluation, costs, Docker/K8s, Responsible AI, and security.

• LLM observability: traces, spans, metrics, and alerts
• Continuous evaluation: eval pipelines and regression testing
• Cost optimization: semantic caching, routing, and batching
• Docker for LLMs: containers, multi-stage build, optimization
• Open-source LLM deployment: vLLM, Ollama, Triton Inference
• AWS: ECS Fargate, Lambda, SageMaker endpoints, and CloudWatch
• Responsible AI: bias, fairness, transparency, and explainability
• Security and governance: IAM, KMS, CloudTrail, VPC endpoints
• Production safety: guardrails, moderation, and abuse monitoring

Module VI — Final Project / Capstone (2 ECTS)#

Design, implementation, evaluation, and deployment of a complete system: production RAG + autonomous agent orchestrating multiple tools, deployed on cloud, monitored, and with controlled costs.

Deliverables:

1. Architecture document: technical diagram, ADRs, and stack justification
2. Complete RAG pipeline with RAGAS evaluation ≥ 0.75
3. Autonomous agent with ≥ 3 tools and documented LangSmith traces
4. Cloud deployment: public URL, uptime ≥ 99%, P95 latency < 3 s
5. LLMOps dashboard: cost/query, faithfulness, latency, and alerts
6. Cost analysis: actual breakdown and scaling projection
7. AWS AIF-C01 practice exam: 65 questions with review of improvement areas
8. NCA-GENL preparation: five weighted areas, ten topics, flashcards, and 50-question practice exam
9. Live demo (20 min) + code review + technical defense (15 min Q&A)

---

Certifications#

• AWS Certified AI Practitioner (AIF-C01) — covered in modules 1–5. Material at certificaciones/aws-aif-c01/.
• NVIDIA Certified Associate: Generative AI LLMs (NCA-GENL) — covered in modules 2–6. Material at certificaciones/nvidia-nca-genl/.