During the first year Aulafy will operate as an open, free, versioned program. Each update should improve lessons, practices, sources, rubrics, or projects. The objective is to build an increasingly reliable educational base: less scattered content, clearer paths, more reproducible exercises, and stronger evidence that the learner can do real work.
Modular program in Applied Artificial Intelligence
A seven-module open curriculum for learning AI from zero to building, evaluating, and operating real systems. The goal is to make Aulafy a clear, living, practical path for Spanish-speaking learners who need to update their skills without getting lost among hundreds of tools.
The program rule
You do not progress by reading a page. You progress when you can explain the decision, reproduce the practice, break it safely, debug it, verify it, and save evidence. Each module ends with a project that another person can review.
Aulafy method
- 1Quick diagnosis: the learner chooses what they want to achieve today and receives one recommendation, not a catalogue.
- 2Short mission: every block should end in a 20-to-90-minute practice with a visible result.
- 3Verification: the learner saves the prompt, decision, output, errors, and an explanation of what they understood versus copied.
- 4Natural bridge: after each mission, the learner knows whether to continue into small business, coding, web, RAG, agents, or local models.
AI changes too quickly for a static syllabus. The program separates stable foundations from volatile tools and keeps a weekly review loop for models, APIs, prices, security, regulation, benchmarks, real use cases, and useful new practices.
- X: real questions from learners, small businesses, developers, and workers who need to reskill.
- Reddit: practical problems, common errors, user experiences, and technical debates.
- OpenAI, Anthropic, Google, Meta, xAI, and Mistral: official docs, models, APIs, pricing, safety, and product changes.
- Hugging Face: open models, datasets, leaderboards, licenses, quantization, and local AI trends.
- GitHub: educational repos, agent frameworks, RAG, evaluation, MCP, observability, and reproducible examples.
- Official legal sources: privacy, AI regulation, invoicing, education, and compliance for small businesses.
curriculum
Seven training modules
Orientation and first useful win
Absolute beginner / work urgency
Help the learner stop feeling lost, choose one concrete situation, and get a useful first AI win in under 45 minutes without starting with too much theory or too many tools.
Solve a real problem from today —email, summary, decision, study, or planning— and document what AI did, what you reviewed, and whether it saved time.
Exit gate. You can choose an initial tool, ask for one concrete task, review the result, and explain which data should not be entered yet.
Evidence
- Personal situation diagnosis
- First useful task solved
- Prompt and output saved
- Understood versus copied checklist
AI for small businesses and real automation
Freelancers, small teams, and operations
Detect repetitive tasks, measure lost time, and build a first automation with human control, visible costs, and clear rules so AI prepares work without deciding for the business.
Automate one office task —email, reviews, form, summary, or data entry— with n8n or a simpler alternative, keeping human approval before any sensitive action.
Exit gate. You can explain what AI automates, where a person intervenes, which data crosses each tool, and when the workflow should move to self-hosted.
Base courses
Evidence
- Repetitive task map
- First human-in-the-loop workflow
- Estimated time saved
- Cost, data, and limits note
Programming with AI
Early technical / developer / maker
Move from asking for code in chat to directing development agents with persistent context, AGENTS.md or CLAUDE.md, diff review, traces, tests, and security control.
Ship one real repository improvement using AI: project context, branch, change, validation, diff explanation, and cognitive debt review.
Exit gate. You can direct a coding agent, review what changed, detect invented architecture, and explain why the change is maintainable.
Base courses
Evidence
- Repository with README and AGENTS.md
- Small reviewed diff
- Reusable skill or instruction
- Command and secret safety review
Build professional websites with AI
Advanced non-technical / maker / professional
Build and publish a useful website with AI while avoiding generic template output and caring about responsive design, SEO, speed, calls to action, domain, maintenance, and cost.
Publish a website for a business, student, or professional with a clear proposal, visible CTA, mobile review, basic SEO, and improvement plan.
Exit gate. You can publish a real website and explain why it is not just a pretty demo: what problem it solves, how it converts, how it is maintained, and what risks it has.
Evidence
- Published website or demo
- Anti-AI-slop checklist
- Responsive and SEO review
- Variables, domain, analytics, and costs documented
Advanced and safe RAG
Intermediate technical / business
Build systems that retrieve correct and authorized information only, combining chunking, metadata, BM25, embeddings, re-ranking, evaluation, and abstention to avoid elegant but false answers.
Build a small document assistant that answers with citations, measures whether it retrieved the right passage, and blocks information the model is not authorized to see.
Exit gate. You can explain why RAG fails, measure retrieval, protect documents, and decide when SQLite/FTS5 is enough or when embeddings/vector search are needed.
Base courses
Evidence
- Test document corpus
- Ingestion and retrieval pipeline
- BM25/vector/hybrid comparison
- Evaluation of correct, uncertain, and forbidden cases
Controlled agents
Advanced technical
Create agents with tools, context, memory, harness, loop, traces, stop criteria, minimum permissions, and human review so autonomy does not become an expensive bug.
Create an agent or small agent team that solves a real task, records every step, fails in a controlled way, and asks for approval on sensitive actions.
Exit gate. You can operate an agent with budget, sandbox, traces, safe stop, human-in-the-loop, and clear criteria before granting more permissions.
Base courses
Evidence
- Agent with limited tools
- Permission matrix and allowlist
- Minimum traces and evals
- Failure, recovery, and cost plan
Local models and self-hosted AI
Advanced technical / specialization
Learn when to use cloud and when to use local AI, with an honest view of privacy, cost, VRAM, quantization, Ollama, Open WebUI, local RAG, self-hosted n8n, and open models.
Set up a basic self-hosted system —local chat, RAG, or automation with Ollama— and justify which tasks belong local, cloud, or hybrid.
Exit gate. You can choose local, cloud, or hybrid based on evidence, privacy, cost, and quality without promising miracles on insufficient hardware.
Base courses
Evidence
- Realistic hardware table
- Documented local test
- Local/cloud comparison
- System where sensitive data does not leave the machine
Tracks by profile
Non-technical profile
Start with orientation and small business automation. Technical depth appears only when the goal requires it.
Use AI with judgment, automate a first task, and publish a useful solution without losing privacy or human review.
Developer
Start with AI-assisted programming, but review foundations so you do not build on hype.
Ship changes, apps, RAG, and agents with tests, security, documentation, and deployment.
Engineering / MLOps
Start with data, RAG, agents, and operations while keeping evaluation and cost in the first prototype.
Operate AI systems with observability, evaluation, model routing, security, and budget control.