INTERMEDIATE · 8 MODULES · 64 LESSONS
Architecture for AI-Native Systems
Design systems with an AI component without letting the model touch money or critical data: latency and cost budgets, evals as a gate, guardrails, and fallbacks.
- For you if
- You already hooked an LLM into your system and now face latency, cost, or hallucinations.
- By the end
- You'll know where AI fits in your architecture, how to contain it with deterministic rules, and what to do when it fails.
- You need
- Basic Python and backend experience. The model is simulated, so no API keys needed.

Curriculum
01Qué cambia cuando un componente es no determinista
- Module introduction: what changes when a component is non-deterministic
- The probabilistic contract
- The LLM is a component, not the system
- It's not "just an API call"
- The probabilistic core and the deterministic shell
- How much non-determinism can a feature tolerate?
- The property sheet of an AI component
- Mini-project: place an AI feature into Mercado
02Latencia y costo como arquitectura
- Module introduction: latency and cost as architecture
- The LLM is slow and costs per call
- The latency and cost budget
- The model cascade: cheap first, escalate only if needed
- Cache: many queries repeat
- Async and streaming: when the user can't wait
- The cost-aware request path
- Project: design the budget for a Mercado AI feature
03El eval como fitness function
- Module introduction: the eval as a fitness function
- Why you can't assert a probabilistic component
- The eval-set and the score
- The eval as a quality gate (fitness function)
- Catching regressions when the prompt or model changes
- The eval in CI: a gate on the deploy
- Types of eval criterion
- Project: build the eval gate for a Mercado AI feature
04Guardrails y la frontera de confianza
05Modos de fallo y resiliencia para IA
- Module introduction: failure modes and resilience for AI
- The new failure modes of an AI component
- Hallucination as a failure mode
- The model down, slow, or rate-limited
- Fallback and graceful degradation
- The circuit breaker over the model
- Drift: the silent degradation
- Project: make a Mercado AI feature resilient
06La cáscara determinista
07El lazo de datos y de retroalimentación
- Module introduction: the data and feedback loop
- The flywheel: usage becomes a better system
- Observability for AI: beyond the server log
- The feedback loop as a design decision
- Closing the loop: the feedback becomes the eval-set
- Prompt vs RAG vs fine-tune as an architectural choice
- Routing the feedback signal to the right lever
- Project: build the data loop for a Mercado AI feature
08Proyecto — arquitecta una feature de IA en Mercado
Reviews
What students say
These reviews are from enrolled students who completed at least 50% of the course. We moderate reviews only on content grounds (spam, offensive language, personal data), never for being critical or negative.
No approved reviews yet.
Be the first to share your experience!