Learn from one working project: the main study guide uses OpsAtlas AI throughout. Its RAG, MCP, analytics, workflow, and evaluation sections include short excerpts of the real code with plain-English explanations, so you can study the concept and implementation together.
A portable, plain-language reference for learning how modern AI applications work. Use it to study on the go, prepare for interviews, or quickly revisit a concept without searching through scattered notes and screenshots.
The guide is written from an analytics perspective: how to turn governed data, business definitions, and operational metrics into reliable AI decision-support tools.
Open the complete study guide for a concise overview of the full applied AI system.
Then use the modules when you want to move from understanding to application. The study guide is the concise reference; each module adds a worked example, design decisions, common failure modes, and a practical exercise.
| Topic | What you will understand | Study page |
|---|---|---|
| Foundations | Models, tokens, context windows, temperature, and system boundaries | Foundations |
| Prompts and context | How instructions and supplied information shape an answer | Prompts and context |
| Tools and structured output | How an AI system returns dependable data and calls software | Tools and structured output |
| Embeddings and RAG | How systems search trusted information and answer with evidence | RAG |
| MCP | How AI applications connect to tools and data through a common protocol | MCP |
| Agents | How models, tools, state, routing, and approvals form a workflow | Agents |
| Evaluation | How to measure quality before and after release | Evaluation |
| Security and governance | How to protect data, constrain actions, and preserve accountability | Security |
| Production and monitoring | How to operate a dependable AI application | Production |
| Resource | Use it for | What it contains |
|---|---|---|
| Complete study guide | Fast learning and interview refreshers | Definitions, diagrams, comparisons, examples, and key resources in one page |
| Topic modules | Deeper study and hands-on practice | Worked scenarios, implementation choices, failure analysis, exercises, and completion criteria |
| Roadmap | Planning your learning sequence | Suggested order, milestones, and portfolio deliverables |
flowchart LR
U[User question] --> I[Instructions and context]
I --> M[Model]
M -->|needs knowledge| R[Retrieval / RAG]
M -->|needs an action| T[Tools / MCP]
R --> M
T --> M
M --> G[Guardrails and approvals]
G --> A[Answer or action]
A --> E[Evaluation and monitoring]
The model is one component. A useful AI product also needs trusted context, controlled tools, predictable outputs, evaluation, and operational safeguards.
For each topic:
- Read the plain-language overview and mental model.
- Explain it aloud without using technical jargon.
- Work through the analytics example.
- Answer the review questions from memory.
- Follow the official references when you want implementation detail.
- Build one small example and record what you learned in LEARNING_LOG.md.
The material leads toward an Operational Intelligence Agent that:
- combines global and regional metrics in one view;
- uses Python to create a governed JSON data layer;
- retrieves definitions and operating guidance with citations;
- uses controlled tools to identify the dimensions driving a metric change;
- explains results through an agent, custom GPT, or HTML dashboard;
- evaluates answer quality, tool use, latency, and cost;
- requires human approval for consequential actions.
This connects existing analytics experience with applied AI: the goal is faster, traceable diagnosis and better decisions, rather than a chatbot for its own sake.
- OpenAI developer learning resources
- OpenAI Agents guide
- OpenAI Agents SDK
- OpenAI tools guide
- OpenAI MCP guide
- Model Context Protocol documentation
- LangChain learning resources
- LangSmith evaluation concepts
Asteway Kebede — Data Analytics & Applied AI
LinkedIn · GitHub