AI And Data Architecture Terms
AI, retrieval, data, evaluation, and LLM application patterns that recur in FDE listings.
Agentic Workflow
A workflow where an LLM-powered system can plan or perform steps using tools, context, state, and sometimes memory. In practical FDE work, this should usually include human review for important actions.
Generative AI Patterns
Common ways of using generative models inside software, such as summarization, classification, extraction, drafting, RAG, tool calling, and workflow recommendations.
Prompt Engineering
The practice of designing instructions, context, examples, constraints, and output formats so an AI system performs a task more reliably.
Agentic AI Orchestration
The coordination layer for an agentic workflow: choosing steps, passing state, calling tools, handling retries, routing to humans, and deciding when the workflow is complete.
Agentic AI Guardrails
Controls that keep an agentic system inside acceptable boundaries, such as permissions, input validation, output validation, policy checks, human approval, audit logs, and safe failure behavior.
RAG
Retrieval-augmented generation. A system retrieves relevant source material and gives it to an LLM so the answer is grounded in specific documents, tickets, data, or records.
Embeddings
Numeric representations of text or other content used to compare semantic similarity. Commonly used for search and retrieval.
Vector Databases
Databases optimized for storing embeddings and finding similar records. Useful for semantic search, though many real systems also need metadata filtering, permissions, freshness checks, and ordinary SQL data.
AI Evaluation / Evals
The practice of measuring whether an AI system performs acceptably for a task. Evals can include golden datasets, human review, automated checks, regression tests, safety checks, and production monitoring.
Data Ingestion
The process of collecting data from source systems and loading it into an application, warehouse, index, or pipeline. In FDE work, ingestion often includes permissions, schema mapping, freshness, validation, and failure handling.
Data Pipeline
A repeatable flow that moves and transforms data from sources to destinations. Pipelines may batch, stream, clean, enrich, validate, join, or index data for downstream applications and analytics.