Phase 5: Cloud + AI Integration (Weeks 21–28)
Objective: Integrate cloud and AI knowledge into complete enterprise solutions. Build RAG systems, learn model evaluation, and earn AI-900 certification (with optional advanced AI-103 path).
🎯 Expected Outcomes
By completing this phase:
- Azure AI Foundry mastered (models, evaluations, agents)
- RAG (Retrieval-Augmented Generation) implemented
- AI evaluation and monitoring in production
- AI-900 certification earned
- Functional RAG system in portfolio
Weeks 21-22: Azure AI Foundry
Objective
Master Microsoft's AI platform for building and deploying enterprise solutions.
Main Resources
| Resource | Language | Type | Duration |
|---|---|---|---|
| Azure AI Foundry Documentation | 🇪🇸 Spanish | Docs | 8 hrs |
| Microsoft Learn: Azure AI Foundry (What is it?) | 🇪🇸 Spanish | Docs | 6 hrs |
| Azure AI Foundry Quickstarts | 🇪🇸 Spanish | Labs | 4 hrs |
| Azure OpenAI Service | 🇪🇸 Spanish | Docs | 4 hrs |
| DeepLearning.AI: Building with Azure AI | 🇬🇧 English | Course | 3 hrs |
Week 21 Plan
| Day | Topic | Practice |
|---|---|---|
| Monday | Azure AI Foundry: overview, projects, hubs | Create project |
| Tuesday | Available models: GPT-4, GPT-4o, Phi | Compare models |
| Wednesday | Playground: prompt testing and refinement | Experiment |
| Thursday | Deployments: quotas, rate limits, regions | Deploy model |
| Friday | Content filters and safety settings | Configure safety |
Week 22 Plan
| Day | Topic | Practice |
|---|---|---|
| Monday | Azure AI Agents: overview and types | Explore Agent Service |
| Tuesday | Agents with tools: code interpreter, file search | Build agent |
| Wednesday | Evaluations: built-in and custom metrics | Evaluate responses |
| Thursday | Prompt flow: orchestration flows | Design flow |
| Friday | Monitoring and logging in production | Metrics dashboard |
Key Concepts
Azure AI Foundry:
├── Hub (organización)
│ └── Project (workspace)
│ ├── Models
│ │ ├── Azure OpenAI (GPT-4, GPT-4o)
│ │ ├── Open models (Phi, Llama, Mistral)
│ │ └── Custom fine-tuned
│ ├── Agents
│ │ ├── Code Interpreter
│ │ ├── File Search
│ │ └── Custom Tools
│ ├── Evaluations
│ │ ├── Groundedness
│ │ ├── Relevance
│ │ ├── Coherence
│ │ └── Custom metrics
│ └── Deployments
│ ├── Serverless
│ └── Provisioned
Weeks 23-24: RAG (Retrieval-Augmented Generation)
Objective
Build a system that answers questions based on internal documents, eliminating hallucinations.
Main Resources
| Resource | Language | Type | Duration |
|---|---|---|---|
| Microsoft Learn: RAG with Azure AI | 🇪🇸 Spanish | Docs | 3 hrs |
| DeepLearning.AI: Building RAG Applications | 🇬🇧 English | Course | 3 hrs |
| LangChain RAG Tutorial | 🇬🇧 English | Tutorial | 4 hrs |
| Azure AI Search Documentation | 🇪🇸 Spanish | Docs | 4 hrs |
| Pinecone: RAG Guide | 🇬🇧 English | Guide | 2 hrs |
Week 23 Plan
| Day | Topic | Practice |
|---|---|---|
| Monday | RAG concepts: why, how, architecture | Diagram system |
| Tuesday | Embeddings: text → vectors | Generate embeddings |
| Wednesday | Vector stores: Azure AI Search, Pinecone | Setup vector store |
| Thursday | Chunking strategies: size, overlap | Process documents |
| Friday | Retrieval: similarity search, hybrid | Implement search |
Week 24 Plan
| Day | Topic | Practice |
|---|---|---|
| Monday | Generation: prompt with context | System prompts for RAG |
| Tuesday | Evaluation: groundedness, relevance | Measure quality |
| Wednesday | Advanced: reranking, metadata filters | Improve precision |
| Thursday | Production: caching, monitoring | Optimize |
| Friday | Deploy complete project | — |
🔨 Project: Internal Policy Chatbot
Chatbot that answers questions about an organization's internal policies:
Policy Chatbot Architecture:
├── Document Ingestion
│ ├── PDF/Word processing
│ ├── Chunking (500 tokens, 50 overlap)
│ └── Embedding generation
├── Vector Store
│ ├── Azure AI Search / ChromaDB
│ └── Metadata indexing
├── Retrieval
│ ├── Similarity search (top-k)
│ ├── Hybrid search (vector + keyword)
│ └── Reranking
├── Generation
│ ├── System prompt (compliance expert)
│ ├── Context injection
│ └── Source citations
└── Interface
├── Streamlit UI
└── API endpoint
Example documents:
- Information security policy
- Acceptable use policy
- Incident response plan
- Data privacy policy
Deliverable: Complete repository with code, documentation, and demo.
Weeks 25-26: AI Monitoring and Safety
Objective
Learn to monitor AI systems in production and detect issues before they impact users.
Main Resources
| Resource | Language | Type | Duration |
|---|---|---|---|
| Microsoft: AI Content Safety | 🇪🇸 Spanish | Docs | 3 hrs |
| Azure AI Evaluation SDK | 🇪🇸 Spanish | Tutorial | 4 hrs |
| Giskard: AI Testing | 🇬🇧 English | Tool | 3 hrs |
| OWASP Top 10 for LLMs | 🇬🇧 English | Framework | 4 hrs |
| Microsoft: Red Teaming AI | 🇪🇸 Spanish | Guide | 2 hrs |
Week 25 Plan
| Day | Topic | Practice |
|---|---|---|
| Monday | Hallucinations: causes and detection | Evaluate models |
| Tuesday | Content safety: filters and moderation | Implement filters |
| Wednesday | Prompt injection: attacks and defenses | Red team exercise |
| Thursday | Data leakage: prevention | PII detection |
| Friday | OWASP Top 10 for LLMs | Assessment |
Week 26 Plan
| Day | Topic | Practice |
|---|---|---|
| Monday | Evaluation metrics: groundedness, coherence | Evaluation SDK |
| Tuesday | A/B testing prompts and models | Testing framework |
| Wednesday | Monitoring dashboards | Azure Monitor + AI |
| Thursday | Alerting and incident response for AI | Playbooks |
| Friday | Complete AI Red Team exercise | Document findings |
🔨 Project: AI Safety Testing Framework
Create a testing framework that includes:
- Test cases for prompt injection
- Groundedness evaluation (hallucinations)
- PII detection checks
- Content safety validation
- Performance benchmarks
- Reporting template
Weeks 27-28: AI-900 Certification (+ Optional AI-103)
Objective
Consolidate knowledge and earn AI-900 certification.
AI-900 Resources
| Resource | Language | Type | Duration |
|---|---|---|---|
| Microsoft Learn: AI-900 | 🇪🇸 Spanish | Learning Path | 8 hrs |
| AI-900 Study Guide | 🇪🇸 Spanish | Official guide | Reference |
| John Savill's Technical Training (YouTube) | 🇬🇧 English | Video (study cram) | 2 hrs |
| Official AI-900 Practice Assessment (free) | 🇪🇸 Spanish | Free official practice | 2 hrs |
Week 27 Plan
| Day | Module | Topic |
|---|---|---|
| Monday | Module 1 | AI workloads and considerations |
| Tuesday | Module 2 | Machine Learning fundamentals |
| Wednesday | Module 3 | Computer Vision |
| Thursday | Module 4 | NLP and Document Intelligence |
| Friday | Module 5 | Generative AI |
Week 28 Plan
| Day | Activity |
|---|---|
| Monday | General review + gaps |
| Tuesday | Practice exams |
| Wednesday | Review weak areas |
| Thursday | Final practice exam |
| Friday | AI-900 exam |
| Saturday | If passed: plan next credential (AI-103 or vendor-neutral) |
Certification: AI-900
- Cost: $99 USD
- Language: Available in Spanish
- Format: 40-60 questions, 45 minutes
- Passing score: 700/1000
- Domains:
- AI workloads and considerations (15-20%)
- Machine Learning on Azure (20-25%)
- Computer Vision workloads (15-20%)
- NLP workloads (15-20%)
- Generative AI workloads (15-20%)
Optional: AI-103 (if there is time and motivation)
For advanced candidates with a development objective, AI-103 adds value:
- Associate level (more weight than Fundamentals)
- Successor to AI-102 (retired Jun 30, 2026); covers generative AI apps and multi-agent solutions with Microsoft Foundry (Python)
- Verify price, status (beta/GA), and syllabus on the official AI-103 page
- Requires 4-6 additional weeks of preparation
If your target is AI Program Manager / AI Governance / TPM (not writing production code), a vendor-neutral credential such as IAPP AIGP or PMI-CPMAI (exam in Spanish) often provides more return than AI-103. See the full ladder in Plan Overview.
The patterns in this phase (RAG, evaluation, agents, safety) do not depend on one vendor. Everything you build in Azure AI Foundry can be replicated with LangChain / LlamaIndex, open models (Llama, Mistral, Phi), and open vector stores (ChromaDB, pgvector, FAISS). Use Azure as one example, but document your project in a vendor-agnostic way so it applies to employers using AWS, GCP, or open-source stacks.
📋 Phase 5 Checklist
- Azure AI Foundry: models, agents, evaluations
- RAG: complete system implemented
- Policy Chatbot: functional and documented
- AI Safety: OWASP LLM Top 10 understood
- Red teaming: exercise completed
- AI Safety Testing Framework: created
- AI-900 Learning Path: completed
- AI-900 certification: earned
- Portfolio: 10+ projects in GitHub
🔗 Resume Value
After this phase:
"I implemented enterprise RAG systems with Azure AI Foundry, including document ingestion, vector search, and generation with source citations. I designed AI safety testing frameworks that evaluate hallucinations, prompt injection, and data leakage. Certifications: AZ-900, SC-900, AI-900."