AI-102 Exam Prep
AI-102 Exam Glossary - 37 Terms
Search the terminology pack for Designing and Implementing a Microsoft Azure AI Solution. Use these definitions with the study guide and practice questions.
A
- Azure AI Custom Vision Object Detection
- A computer vision capability used to identify and locate objects in images by returning bounding boxes for each detected instance.
- Azure AI Video Indexer
- A service that analyzes video and audio content to generate searchable insights such as transcripts, speakers, topics, and brand mentions.
- Azure Translator Document Translation API
- An Azure translation feature used to translate entire documents while preserving their original formatting and structure.
B
- Basic tier
- An Azure AI Search pricing tier that does not support certain advanced capabilities such as semantic search.
- BLEU score
- A machine translation evaluation metric that compares generated translations against reference translations.
- Bounding box
- A rectangular coordinate region that marks the location of a detected object within an image.
- Brand mention detection
- The process of identifying references to company or product brands within media content.
C
- CLU
- Conversational Language Understanding, an Azure service used to detect intents and extract entities from user utterances.
- Custom Neural model
- A custom document intelligence model trained to extract fields from documents with varied layouts and formats.
- Custom Speech
- An Azure Speech capability that improves speech recognition accuracy for domain-specific vocabulary by training with custom data.
- Custom Translator
- A feature for training translation models with domain-specific bilingual data to improve terminology and translation quality.
- Custom Web API skill
- A skill in an Azure AI Search skillset that calls an external REST endpoint during document enrichment.
E
- Embedding field
- A vector-based field in a search index that stores numerical representations of content for similarity search.
F
- File projection
- A knowledge store projection type used to store binary outputs such as images extracted during enrichment.
G
- Groundedness
- An evaluation measure indicating whether a generated response is supported by and faithful to the provided source content.
I
- Intent
- The goal or purpose a user is expressing in a natural language request.
K
- Knowledge store
- A persistence layer in Azure AI Search that stores enriched content generated during indexing for downstream analysis.
O
- OCR
- Optical Character Recognition, a process that extracts readable text from images or scanned documents.
- Overfitting
- A modeling problem in which a system performs well on training data but poorly on unseen real-world inputs.
P
- Parallel data
- Aligned source and target language sentence pairs used to train or improve machine translation models.
- Projection
- A configuration in Azure AI Search knowledge store that shapes enriched data into tables, objects, or files for storage.
R
- RAG
- Retrieval-Augmented Generation, an architecture that combines retrieved source documents with a generative model to produce answers.
- Regex entity
- A pattern-based entity type used to extract values that match a consistent textual format, such as account numbers.
- Regular expression
- A formal pattern syntax used to match, validate, or extract structured text strings.
- REST API endpoint
- A web-accessible service interface used by applications or skills to send requests and receive responses over HTTP.
S
- Semantic search
- A search capability that uses language understanding to improve relevance beyond simple keyword matching.
- Semantic similarity
- The degree to which two pieces of content are related in meaning rather than exact wording.
- Skillset
- A collection of enrichment steps in Azure AI Search that process content during indexing.
- Speaker identification
- The capability to distinguish and label different speakers in audio or video content.
- Speech recognition
- The process of converting spoken language into text.
- Standard S1 tier
- An Azure AI Search service tier that supports semantic search and other more advanced features.
T
- Table projection
- A knowledge store projection type used to store structured enrichment outputs in tabular form for querying and reporting.
- Text Merge
- An enrichment step that combines OCR-extracted text with existing document text so all content becomes searchable.
- Topic detection
- A technique for identifying the main subjects discussed in text, audio, or video content.
- Transcript extraction
- The process of converting spoken audio in media into text for search, analysis, or indexing.
U
- Utterance
- A sample phrase or sentence used to train or evaluate language understanding models.
V
- Vector search
- A search approach that retrieves content based on similarity between vector embeddings rather than exact keyword matches.
About These Definitions
These definitions are loaded from the shared release pack. Use them with the study guide and practice questions to connect vocabulary to exam scenarios.