RAG Specialist

The only genuinely non-negotiable prerequisite in the entire Atlas.

The specific mathematics that appears in practice, and nothing beyond it.

Vectors, matrices and the operations every model is secretly made of.

Derivatives, the chain rule, and what it means to walk downhill in a loss landscape.

Systems whose behaviour is learned from data rather than written as rules.

The optimisation procedure underneath essentially every model in this Atlas.

Many-layered neural networks that learn their own features instead of being handed them.

Layers, weights and activations. The unit of construction for everything downstream.

Letting every position look at every other position. The hinge the modern field turns on.

Attention, feed-forward layers, residuals and normalisation. The block that ate the field.

Transformers trained on enough text to become general-purpose reasoning surfaces.

Meaning as coordinates. The representation that makes search, clustering and RAG possible.

Give the model the right documents at query time instead of hoping it memorised them.

Where embeddings live: indexes, filters, and the operational reality of them.

Approximate nearest neighbours at scale. HNSW, IVF, and the recall/latency trade.

A second, slower pass over the top results. Usually the largest single quality win available.

Distributions, expectation, and why every evaluation number needs an error bar.

Learning a mapping from labelled examples. The workhorse of applied ML.

The skill that separates people who ship models from people who publish notebooks.

Measuring quality when there is no single correct output. The hardest unsolved problem in shipping.

How you split documents. Where most RAG systems quietly lose their quality.

Separating retrieval failures from generation failures, so you fix the right one.

Combining keyword and semantic retrieval, because each fails where the other works.

Ingestion pipelines, freshness, permissions, caching and cost at real volume.