Skip to content
AI Atlas

AI Engineering, Safety & Ethics

Turning a working model into a usable, trustworthy, accountable product

OVERVIEW

All Concepts
6
Beginner
2
Intermediate
3
Expert
1

Getting a model to work in a notebook is roughly a third of the job. The rest is engineering and governance: compressing it to a deployable cost, evaluating it rather than trusting a feeling, defending against injection and misuse, and being able to explain and attribute failure when it happens. This domain is the gate every model must pass to become a product.

Questions this domain answers

  • Q1

    How are large models compressed onto consumer hardware?

  • Q2

    How do you evaluate when there is no answer key?

  • Q3

    Why is prompt injection so hard to eliminate?

Concepts in this domain

  1. 01Training & Inference InfrastructureExpertMemory decides how large a model you can train, communication how long it takes — raw compute is rarely the bottleneck
  2. 02Model CompressionIntermediateMake a model smaller, faster and cheaper with almost no accuracy loss — but you can usually have only two of the three at once
  3. 03Inference Optimization & ServingIntermediateTraining happens once; inference happens a billion times a day — and serving is torn between fast first tokens and high throughput, which usually pull against each other
  4. 04Retrieval-Augmented GenerationBeginnerRather than cramming knowledge into parameters, leave it outside and look it up on demand — an open-book exam instead of a closed-book one
  5. 05Agents & Tool UseIntermediateLet a model do more than answer: search, call APIs, run code — and decide the next step from what came back
  6. 06Safety, Alignment & Prompt InjectionBeginnerA model optimises the proxy we wrote into the loss, never the thing we actually want — the gap between them is the whole alignment problem