AI Systems Engineering — RAG, agents and evaluation

University pathway · 72 hours

Build, measure and defend an AI system with RAG, memory, orchestration, local models and real evaluation.

What you will learn

The 10-unit engineering roadmap

Every unit has one plain-language explanation, one technical decision and one piece of evidence for your final system.

Final capstone

Defend a RAG system that knows when not to answer.

Deliver the ingestion plan, retrieval evaluation, citations, memory policy, cost model, red-team results and the limits you would disclose to a real user.

Model foundations

Tokens, transformers, training, inference, SLMs, frontier models and GPU constraints.

Grounded RAG

Documents, chunking, embeddings, reranking, citations and abstention when evidence is missing.

Reliable systems

Memory, agents, human approval, diffusion, multimodality, evaluation and red teaming.

72 guided hours

10 engineering units

40 lessons & labs

1 defended capstone

Tokens & transformers

Context, cost and representation

Training & inference

Choose prompting, RAG or fine-tuning

SLMs, frontier & open source

Quality, privacy, licence and control

GPU, VRAM & serving

Latency, capacity and cost

Retrieval & embeddings

Documents, chunks and metadata

Reranking & citations

Evidence before fluent answers

Memory & context

Retention, correction and deletion

Orchestration & agents

Tools, state and human approval

Diffusion & multimodality

Images, audio, rights and tests

Evaluation & red teaming

Measure failure before users find it