Phase 2: Technical Portfolio — RAG, Agents, and Evaluation (Weeks 5–12)
Objective: Build the core technical evidence for the AI Solution Architect profile. In 8 weeks you produce 3 strong, reproducible artifacts (a data pipeline with business value, an evaluated RAG system, and an agent comparison) that support the architecture narrative in interviews.
🎯 Expected Outcomes
By completing this phase:
- Data pipeline feeding an AI use case, with documented impact metric
- RAG system implemented with reproducible evaluation (groundedness, relevance)
- Agent comparison with traces and selection criteria
- 3 repos with professional README files ready for Gate CP2 (Week 12)
"Proof over study": every week closes with published evidence (repo, benchmark, or write-up). See Methodology.
Block A — Data that feeds AI (Weeks 5–8)
Week 5 — Data project #1: ingestion and cleaning
Objective: build a reproducible pipeline that transforms raw data into an AI-ready dataset (the foundation for a RAG or dashboard).
| Resource | Language | Type |
|---|---|---|
| Python: pandas | 🇬🇧 | Docs |
| Microsoft Learn: Power Query | 🇪🇸 | Learning Path |
| SQL — SQLBolt | 🇬🇧 | Interactive |
Deliverable: repo data-pipeline/ — ingestion + cleaning script, versioned output dataset, README with flow diagram and one metric (records processed, % errors corrected).
Week 6 — Data project #2: impact metric
Objective: enrich the pipeline with a second source and calculate a business metric (time savings, coverage, quality).
Deliverable: notebook or script that produces the before/after metric + short write-up explaining "why this matters to the business."
Week 7 — Executive dashboard + write-up
Objective: communicate the result to an executive audience.
| Resource | Language | Type |
|---|---|---|
| Microsoft Learn: Power BI | 🇪🇸 | Learning Path |
| DAX Guide (SQLBI) | 🇬🇧 | Reference |
Deliverable: dashboard (Power BI or similar) with 3–5 KPIs + 1-page write-up with the data story.
Week 8 — Workflow automation + README
Objective: remove manual steps from the pipeline.
| Resource | Language | Type |
|---|---|---|
| Power Automate | 🇪🇸 | Learning Path |
| Python: scheduling / basic cron | 🇬🇧 | Docs |
Deliverable: documented automated flow + README anyone can run. Block A close: publishable artifact #1.
Block B — RAG and Agents (Weeks 9–12)
Week 9 — RAG prerequisite lab + baseline benchmark
Objective: understand RAG architecture and establish a measurable quality baseline.
| Resource | Language | Type |
|---|---|---|
| Microsoft Learn: RAG with Azure AI | 🇪🇸 | Docs |
| LangChain: RAG Tutorial | 🇬🇧 | Tutorial |
| Azure AI Search | 🇪🇸 | Docs |
Deliverable: RAG architecture diagram + a set of 15–20 evaluation questions (your "baseline benchmark").
Week 10 — RAG implementation + evaluation
Objective: build the RAG and measure its quality (not just that it "works").
| Resource | Language | Type |
|---|---|---|
| Azure AI Evaluation SDK | 🇪🇸 | Tutorial |
| Chunking strategies | 🇪🇸 | Docs |
Reference architecture: ingestion → chunking → embeddings → vector store (Azure AI Search / Chroma) → retrieval (hybrid + reranking) → generation with source citations → evaluation (groundedness, relevance).
Deliverable: repo rag-evaluado/ — functional RAG + evaluation report with metrics per question. Publishable artifact #2.
Week 11 — Comparative agents + traces
Objective: compare 2–3 agent approaches on the same task and justify the selection.
| Resource | Language | Type |
|---|---|---|
| Azure AI Agent Service | 🇪🇸 | Docs |
| Agent Framework: Your First Agent | 🇬🇧 | Lab |
| LangGraph (open, portable) | 🇬🇧 | Docs |
Deliverable: repo with the same task solved by 2–3 agents, captured execution traces, and a trade-off table (latency, cost, reliability).
Week 12 — CP2 close: 3 strong artifacts
Objective: make the three artifacts hiring-ready and pass the gate.
Checklist for Gate CP2:
- 3 repos with professional README files
- At least 1 reproducible benchmark (the evaluated RAG)
- At least 1 published technical write-up
- Architecture whiteboard narrative ready
📋 Phase 2 Checklist
- Data pipeline with impact metric (artifact #1)
- RAG system with reproducible evaluation (artifact #2)
- Agent comparison with traces (artifact #3)
- Executive dashboard + write-up published
- 3 repos with professional README files
- Gate CP2 approved
v4 Operations
🔗 Resume Value
"I built an enterprise RAG system with reproducible evaluation (groundedness/relevance), an agent comparison with traces and selection criteria, and a data pipeline with measurable business impact — all published on GitHub."