Skip to content

Implement Custom Language Models

This section of the Microsoft AI-102: Designing and Implementing a Microsoft Azure AI Solution exam covers building and deploying custom language models with Azure AI Language and related services. Below are study notes for each sub-topic, with links to Microsoft documentation, exam tips, and key facts

Create Intents, Entities, and Add Utterances

📖 Docs: Language Understanding (LUIS) migration to Conversational Language Understanding

Overview

  • Intents = user goals (e.g., "BookFlight")
  • Entities = data extracted (e.g., "Paris" as destination)
  • Utterances = example phrases to train model

Key Points

  • Provide diverse utterances to improve accuracy
  • Entities can be prebuilt (dates, numbers) or custom
  • Used in chatbots, assistants, and task automation

Train, Evaluate, Deploy, and Test a Language Understanding Model

📖 Docs: Quickstart: Conversational language understanding

Overview

  • Training uses labeled intents and entities
  • Evaluation ensures performance
  • Deployment makes model available via endpoint

Key Points

  • Metrics: precision, recall, F1 score
  • Deploy models to staging or production slots
  • Test with new utterances for real-world accuracy

Exam Tip

Staging slot for testing, production slot for live apps

Optimize, Backup, and Recover Language Understanding Model

📖 Docs: Back up and recover your conversational language understanding models

Overview

  • Optimization involves retraining with new utterances
  • Backup ensures models can be restored
  • Recovery allows rollback after errors

Key Points

  • Export/import projects for portability
  • Regular retraining improves accuracy
  • Versioning helps track model changes

Consume a Language Model from a Client Application

📖 Docs: Quickstart: Conversational language understanding

Overview

  • Applications call deployed models via REST API or SDKs
  • Input: user utterance
  • Output: intent + entities + confidence score

Key Points

  • Requires endpoint URL and key
  • Supports real-time integration into bots and apps
  • JSON response can trigger business logic

Use Case

Chatbot calling CLU to extract intent and act on it

Create a Custom Question Answering Project

📖 Docs: Custom Question Answering overview

Overview

  • Custom QnA project enables building a knowledge base
  • Uses semi-structured or unstructured data as input

Key Points

  • Sources: FAQ docs, URLs, PDFs
  • Knowledge base provides structured Q&A pairs
  • Supports conversational interaction

Add Question-and-Answer Pairs and Import Sources

📖 Docs: What is custom question answering?

Overview

  • Add manual Q&A pairs
  • Import FAQs from documents or URLs
  • Supports multiple sources per knowledge base

Key Points

  • Automatic extraction generates suggested Q&A pairs
  • Manual editing improves accuracy

Train, Test, and Publish a Knowledge Base

📖 Docs: Create, test, and deploy: CQA knowledge base

Overview

  • Training refines extracted Q&A pairs
  • Testing validates accuracy
  • Publishing exposes an endpoint

Key Points

  • Knowledge base must be published to be consumed
  • Endpoint integrates with client applications
  • Supports iterative improvement

Create a Multi-Turn Conversation

📖 Docs: Add guided conversations with multi-turn prompts

Overview

  • Multi-turn = follow-up questions within context
  • Supports contextual conversations

Key Points

  • Requires linking related Q&A pairs
  • Improves chatbot interactivity
  • Maintains conversation state

Add Alternate Phrasing and Chit-Chat to a Knowledge Base

📖 Docs: Improve quality of response with synonyms

Overview

  • Alternate phrasing improves recognition of user queries
  • Chit-chat adds casual responses to non-task queries

Key Points

  • Reduces fallback to “no answer”
  • Improves conversational flow
  • Prebuilt chit-chat sets available

Export a Knowledge Base

📖 Docs: Create, test, and deploy: CQA knowledge base

Overview

  • Knowledge bases can be exported to JSON
  • Supports backup, migration, and versioning

Key Points

  • Importing restores or replicates KBs
  • Useful for moving between environments

Create a Multi-Language Question Answering Solution

📖 Docs: Create projects in multiple languages

Overview

  • Projects can contain multilingual content

Key Points

  • Detect language first, then route query
  • Translator can pre-process inputs
  • Requires maintaining localized KBs

Implement Custom Translation

📖 Docs: Custom Translator

Overview

  • Custom Translator allows domain-specific translations
  • Supports training with parallel text corpora

Key Points

  • Improves accuracy for industry terms
  • Custom models must be trained and published
  • Integrated with Azure Translator API

Exam Tip

Translator = general use, Custom Translator = domain-specific

Quick‑fire revision sheet

  • 📌 Intents = goals, Entities = extracted data, Utterances = examples
  • 📌 CLU models trained, evaluated, deployed to endpoints
  • 📌 Metrics: precision, recall, F1
  • 📌 Custom QnA builds KBs from docs/FAQs
  • 📌 Multi-turn conversations = contextual Q&A
  • 📌 Alternate phrasing + chit-chat improve coverage
  • 📌 Export KB = backup/migration
  • 📌 Multi-language QnA + Custom Translator for global solutions