LLMs & agentsAdvanced
Generative AI
Foundation models on Databricks, vector search for retrieval, agents built with Agent Bricks and the Agent Framework, and how to evaluate them.
0/26|26 written|~45 h
To doIn progressDonePlannedClick a concept to open its cardSwipe the map, tap a concept
Stage 1 · Foundations
Unity Catalog, the governance layerStage 2 · Models
AI PlaygroundFoundation Model APIsUnity Gateway (formerly AI Gateway)Batch inference with ai_queryStructured outputsPaying for a foundation model: tokens, units and reservationsExternal models and model provider servicesStage 3 · Retrieval
Databricks AI Search (formerly Vector Search)Building a RAG pipelineAI Search index types and sync modesStage 4 · Agents
Agents on DatabricksAgent tools as Unity Catalog functionsModel Context Protocol on DatabricksAgent Bricks: Knowledge Assistant and Supervisor AgentDeploy an agent on Databricks AppsModel services on Unity GatewayAgent and MCP services as Unity Catalog securablesGuardrails for generative applicationsAgent memoryEvaluating agentsEvaluation datasets for generative AIHuman feedback on generative AI outputPrompt registryProduction monitoring for GenAI appsGateway inference tables and usage trackingRetrieval before agents, and evaluation before either goes near production. Most failed generative AI projects fail at retrieval quality, not at the model, which is why the RAG concept spends most of its words on chunking and grounding.
Unity Catalog is a prerequisite and not a formality: vector indexes, agent tools and the models they call are all governed objects.
Resources for this path
Courses, books and repos that cover the whole map go here. None have been added yet; per-concept resources appear on each concept page.