update

Author: madiepev

learn-pr/wwl-data-ai/optimize-finetune-agents/1-introduction.yml

@@ -3,7 +3,7 @@ uid: learn.wwl.optimize-finetune-agents.introduction
 title: Introduction
 metadata:
   title: Introduction
-  description: Introduction to advanced fine-tuning decision-making for AI agents, covering method selection, synthetic data strategies, performance assessment, and model lifecycle management.
+  description: Introduction to advanced fine-tuning decision-making for AI agents, covering method selection, agent quality problem recognition, training data preparation, and optimization strategy design.
   author: madiepev
   ms.author: madiepev
   ms.date: 02/25/2026

learn-pr/wwl-data-ai/optimize-finetune-agents/includes/1-introduction.md

@@ -1,18 +1,18 @@
 Your team has deployed an AI agent and validated its performance through structured evaluation. Now you face the next challenge: the agent handles general hiking inquiries well, but customers request specific expertise in niche domains like winter mountaineering safety protocols or ultralight backpacking gear selection. Base model prompting alone doesn't capture the nuanced expertise needed, and you don't have sufficient real-world training data in these specialized areas.
 
-Effective fine-tuning requires making critical design decisions rather than simply following procedural steps. You must evaluate when fine-tuning offers better value than prompt engineering, assess which fine-tuning approach matches your quality and cost constraints, recognize what makes training data effective versus problematic, and determine whether model performance improvements justify deployment to production.
+Effective fine-tuning requires making critical design decisions rather than simply following procedural steps. You must evaluate when fine-tuning offers better value than prompt engineering, assess which fine-tuning approach matches your quality and cost constraints, recognize what makes training data effective versus problematic, and determine whether model performance improvements justify the investment.
 
-**Adventure Works**, an outdoor adventure company, operates a Trail Guide Agent that helps customers plan hiking trips with trail recommendations, accommodation bookings, and gear suggestions. After evaluating their agent's baseline performance, the team identifies opportunities to specialize the model for advanced technical queries about gear specifications and safety equipment recommendations. They need to design a fine-tuning strategy that improves domain expertise without degrading general performance, create synthetic training data that addresses gaps in their limited historical conversation logs, monitor whether fine-tuning delivers measurable quality improvements, and establish a reliable process for deploying fine-tuned models to production.
+**Adventure Works**, an outdoor adventure company, operates a Trail Guide Agent that helps customers plan hiking trips with trail recommendations, accommodation bookings, and gear suggestions. After evaluating their agent's baseline performance, the team identifies opportunities to specialize the model for advanced technical queries about gear specifications and safety equipment recommendations. They need to design a fine-tuning strategy that improves domain expertise without degrading general performance, prepare training data that meets format and quality requirements for their chosen method, and design an optimization strategy that produces measurable improvements.
 
-This module develops your professional judgment for advanced model customization decisions. You learn to evaluate and select appropriate fine-tuning methods based on use case requirements, design synthetic data generation strategies that produce high-quality training data, assess model performance and make optimization decisions using quantitative metrics, and architect model lifecycle workflows from development through production deployment.
+This module develops your professional judgment for advanced model customization decisions. You learn to evaluate and select appropriate fine-tuning methods based on use case requirements, recognize agent quality problems and match them to the right fine-tuning approach, prepare training data that meets format and quality standards, and design optimization strategies using hyperparameter configuration and iterative evaluation.
 
 ## Learning objectives
 
 By the end of this module, you've learned to:
 
-- **Evaluate and select advanced fine-tuning methods** including supervised fine-tuning (SFT), reinforcement learning from human feedback (RLHF), and direct preference optimization (DPO) based on quality requirements, data availability, and cost constraints.
-- **Design and manage synthetic data strategies** for fine-tuning, including assessing data quality requirements, generating diverse training examples, validating synthetic data effectiveness, and managing data versioning.
-- **Assess and optimize fine-tuned model performance** by interpreting evaluation metrics, identifying performance regressions, making data-driven optimization decisions, and validating improvements meet production thresholds.
-- **Architect model lifecycle workflows** from development through production deployment, including model versioning strategies, testing approaches, deployment pipeline design, and rollback procedures.
+- **Evaluate and select fine-tuning methods** including supervised fine-tuning (SFT), reinforcement fine-tuning (RFT), and direct preference optimization (DPO) based on quality requirements, data availability, and cost constraints.
+- **Recognize agent quality problems** and determine which fine-tuning approach addresses the root cause for format inconsistency, tone misalignment, or reasoning failures.
+- **Prepare training data** by validating format requirements for each fine-tuning method and applying quality principles to create effective datasets.
+- **Design an optimization strategy** by evaluating baseline performance, setting measurable targets, splitting your dataset, and configuring hyperparameters for your chosen method.
 
 Let's start by exploring how to evaluate and select fine-tuning methods for your use case.

learn-pr/wwl-data-ai/optimize-finetune-agents/index.yml

@@ -2,7 +2,7 @@
 uid: learn.wwl.optimize-finetune-agents
 metadata:
   title: Optimize and fine-tune AI agents for production
-  description: Learn how to implement advanced fine-tuning methods, create and manage synthetic data, monitor fine-tuned model performance, and manage models from development through production deployment in a generative AI ops workflow.
+  description: Learn how to select fine-tuning methods (SFT, RFT, DPO), recognize agent quality problems, prepare training data, and design optimization strategies using hyperparameter configuration and iterative evaluation.
   author: madiepev
   ms.author: madiepev
   ms.date: 02/25/2026
@@ -13,12 +13,13 @@ metadata:
   ms.collection:
     - wwl-ai-copilot
 title: Optimize and fine-tune AI agents for production
-summary: "Learn to evaluate fine-tuning methods, design synthetic data strategies, assess model performance, and manage model lifecycles from development through production."
+summary: "Learn to evaluate and select fine-tuning methods, recognize agent quality problems, prepare training data, and design optimization strategies using hyperparameter configuration and iterative evaluation."
 abstract: |
   By the end of this module, you'll be able to:
     - Evaluate and select fine-tuning methods including supervised fine-tuning (SFT), reinforcement fine-tuning (RFT), and direct preference optimization (DPO) based on quality requirements, data availability, and cost constraints.
-    - Design and manage synthetic data strategies for fine-tuning, including assessing data quality requirements, generating diverse training examples, validating synthetic data effectiveness, and managing data versioning.
-    - Assess and optimize fine-tuned model performance by interpreting evaluation metrics, identifying performance regressions, making data-driven optimization decisions, and validating improvements meet production thresholds.
+    - Recognize agent quality problems and determine which fine-tuning approach addresses the root cause for format inconsistency, tone misalignment, or reasoning failures.
+    - Prepare training data by validating format requirements for each fine-tuning method and applying quality principles to create effective datasets.
+    - Design an optimization strategy by evaluating baseline performance, setting measurable targets, splitting your dataset, and configuring hyperparameters for your chosen method.
 prerequisites: |
   Before starting this module, you should be familiar with fundamental AI concepts and services in Azure.
 iconUrl: /learn/achievements/generic-badge.svg