You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Welcome to Module 6, which focuses on the design extensibility of AI solutions in enterprise environments. This module introduces solution architects to the foundational concepts, architectural patterns, and best practices for building scalable, secure, and customizable AI solutions using Microsoft platforms. Extensibility is a critical capability that enables organizations to tailor AI systems to their unique business processes, compliance requirements, and operational constraints.
1
+
This module focuses on designing extensible AI solutions in enterprise environments. It introduces solution architects to the foundational concepts, architectural patterns, and best practices for building scalable, secure, and customizable AI solutions by using Microsoft platforms. Extensibility is a critical capability that enables organizations to tailor AI systems to their unique business processes, compliance requirements, and operational constraints.
2
2
3
-
Throughout this module, you explore how to use custom models in Microsoft Foundry, design and operationalize agents within Microsoft 365 Copilot, and extend agent capabilities using Copilot Studio—including advanced integration through the Model Context Protocol (MCP). The unit guide you through structured approaches for model and agent design, integration with enterprise systems, governance, lifecycle management, and professional visualizations that can be adapted for documentation and presentations.
3
+
Throughout this module, you explore how to use custom models in Microsoft Foundry, design and operationalize agents within Microsoft 365 Copilot, and extend agent capabilities by using Copilot Studio, including advanced integration through the Model Context Protocol (MCP). The module guides you through structured approaches for model and agent design, integration with enterprise systems, governance, lifecycle management, and professional visualizations that can be adapted for documentation and presentations.
4
4
5
-
By the end of this module, aren't equipped with expert-level guidance and practical frameworks to architect AI solutions that are not only robust and compliant, but also extensible to meet evolving business needs across diverse scenarios and platforms.
5
+
By the end of this module, you are equipped with expert-level guidance and practical frameworks to architect AI solutions that are not only robust and compliant, but also extensible enough to meet evolving business needs across diverse scenarios and platforms.
### Design extensibility of AI solutions — Summary and key takeaways
2
2
3
-
Module 6 introduces solution architects to the principles and best practices for designing extensible AI solutions in enterprise environments, specifically leveraging Microsoft platforms. The module covers the use of custom models with Microsoft Foundry, agent design and operationalization in Microsoft 365 Copilot, extensibility strategies in Copilot Studio, and advanced integration using the Model Context Protocol (MCP). It emphasizes the importance of scalability, security, compliance, and adaptability in architecting AI systems that address unique organizational needs.
3
+
This module introduces solution architects to the principles and best practices for designing extensible AI solutions in enterprise environments by using Microsoft platforms. The module covers the use of custom models with Microsoft Foundry, agent design and operationalization in Microsoft 365 Copilot, extensibility strategies in Copilot Studio, and advanced integration by using the Model Context Protocol (MCP). It emphasizes the importance of scalability, security, compliance, and adaptability in architecting AI systems that address unique organizational needs.
4
4
5
5
### Key takeaways
6
6
7
7
#### Custom models in Microsoft Foundry
8
8
9
-
- Custom models are essential for scenarios requiring domain-specific reasoning, compliance, data sovereignty, unique workflows, and cost optimization.
10
-
11
-
- Foundry provides tools for data preparation, model training, evaluation, deployment, and governance, supporting full lifecycle management.
12
-
13
-
- Solution architects should follow a structured approach: define business objectives, assess data requirements, select the appropriate model path, integrate with enterprise systems, and validate results.
- Custom models are essential for scenarios requiring domain-specific reasoning, compliance, data sovereignty, unique workflows, and cost optimization.
10
+
- Foundry provides tools for data preparation, model training, evaluation, deployment, and governance, supporting full lifecycle management.
11
+
- Solution architects should follow a structured approach: define business objectives, assess data requirements, select the appropriate model path, integrate with enterprise systems, and validate results.
- Copilot agents are modular, instruction-driven components that automate tasks, retrieve information, and collaborate within Microsoft 365 apps.
20
-
21
-
- Effective agent design centers on clear intent, strict guardrails, scoped permissions, and alignment with business goals and data boundaries.
22
-
23
-
- Collaborative agents can support sequential, parallel, and orchestrated workflows, improving cross-app productivity and user experience.
24
-
25
-
- Lifecycle management includes monitoring agent quality, updating instructions, enforcing access control, and versioning.
16
+
- Copilot agents are modular, instruction-driven components that automate tasks, retrieve information, and collaborate within Microsoft 365 apps.
17
+
- Effective agent design centers on clear intent, strict guardrails, scoped permissions, and alignment with business goals and data boundaries.
18
+
- Collaborative agents can support sequential, parallel, and orchestrated workflows, improving cross-app productivity and user experience.
19
+
- Lifecycle management includes monitoring agent quality, updating instructions, enforcing access control, and versioning.
26
20
27
21
#### Agent extensibility in Copilot Studio
28
22
29
-
- Extensibility is implemented at four layers: instruction-level, skills/capabilities, integration, and pro-code customization via Visual Studio Code.
30
-
31
-
- Modular and reusable agent components enable faster updates, better compliance, and long-term maintainability.
32
-
33
-
- Multi-agent collaboration and domain-context patterns allow agents to specialize and adapt to different business environments.
23
+
- Extensibility is implemented at four layers: instruction-level, skills/capabilities, integration, and pro-code customization through Visual Studio Code.
24
+
- Modular and reusable agent components enable faster updates, better compliance, and long-term maintainability.
25
+
- Multi-agent collaboration and domain-context patterns allow agents to specialize and adapt to different business environments.
34
26
35
27
#### Model Context Protocol (MCP) in Copilot Studio
36
28
37
-
- MCP provides a standardized contract for agents to retrieve and interpret structured business context, ensuring consistent reasoning and compliance.
38
-
39
-
- MCP is particularly valuable for Dynamics 365 Finance & Operations scenarios, enabling agents to access business entities, workflows, and domain models.
- MCP provides a standardized contract for agents to retrieve and interpret structured business context, ensuring consistent reasoning and compliance.
30
+
- MCP is particularly valuable for Dynamics 365 Finance & Operations scenarios, enabling agents to access business entities, workflows, and domain models.
31
+
- MCP-enabled agents improve accuracy, reduce incorrect information, enhance auditability, and support policy-aligned actions across enterprise workflows.
Copy file name to clipboardExpand all lines: learn-pr/wwl/design-extensibility-ai-solutions/includes/2-design-ai-solutions-custom-models-microsoft-foundry.md
+74-61Lines changed: 74 additions & 61 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -17,124 +17,137 @@ Solution architects determine when custom AI models are needed by evaluating bus
17
17
18
18
### Appropriate scenarios
19
19
20
-
**Domain-specific language and reasoning**<br>Industries such as legal, healthcare, engineering, finance, and manufacturing require AI models that understand specialized terminology and follow domain-specific logic.
20
+
#### Domain-specific language and reasoning
21
21
22
-
**High-impact decision processes**<br>Custom models are used when accuracy directly affects compliance, financial outcomes, or operational safety.
22
+
Industries such as legal, healthcare, engineering, finance, and manufacturing require AI models that understand specialized terminology and follow domain-specific logic.
23
23
24
-
**Data sovereignty and governance mandates**<br>Custom models enable organizations to determine exactly how data is processed, stored, evaluated, and monitored.
24
+
#### High-impact decision processes
25
25
26
-
**Unique workflows or personalization requirements**<br>Pre-built copilots may not support custom interaction patterns, long-running processes, or proprietary toolchains.
26
+
Custom models are used when accuracy directly affects compliance, financial outcomes, or operational safety.
27
27
28
-
**Cost optimization for high-volume inference**<br>Small, specialized custom models can provide performance and cost advantages over large foundation models.
28
+
#### Data sovereignty and governance mandates
29
29
30
-
## Understanding architecture foundations in Microsoft Foundry
30
+
Custom models enable organizations to determine exactly how data is processed, stored, evaluated, and monitored.
31
31
32
-
Microsoft Foundry provides an end-to-end platform for custom model development, including tools for data preparation, training, evaluation, deployment, observability, and governance.
32
+
#### Unique workflows or personalization requirements
33
33
34
-
### Key architectural elements
34
+
Pre-built copilots may not support custom interaction patterns, long-running processes, or proprietary toolchains.
35
35
36
-
**Model Catalog**<br>Offers base models that can be fine-tuned or enhanced using enterprise data and specialized tasks.
36
+
#### Cost optimization for high-volume inference
37
37
38
-
**Training and Fine-tuning Pipelines**<br>Help orchestrate data ingestion, labeling, evaluation, and iterative improvements at scale.
38
+
Small, specialized custom models can provide performance and cost advantages over large foundation models.
39
39
40
-
**Agent and Tooling Integration**<br>Custom models can be embedded into Foundry agents and orchestrations to support multi-step reasoning and automated workflows.
40
+
## Understanding architecture foundations in Microsoft Foundry
41
41
42
-
**Responsible AI Controls**<br>Includes content filtering, safety evaluation, transparency artifacts, policy enforcement, and auditability.
42
+
Microsoft Foundry provides an end-to-end platform for custom model development, including tools for data preparation, training, evaluation, deployment, observability, and governance.
43
43
44
-
**Deployment Topologies**
44
+
### Key architectural elements
45
45
46
-
Hosted secure environments
46
+
#### Model catalog
47
47
48
-
Private networking deployments
48
+
Offers base models that can be fine-tuned or enhanced using enterprise data and specialized tasks.
49
49
50
-
Integration with Azure Kubernetes Service and Foundry runtime environments
50
+
#### Training and fine-tuning pipelines
51
51
52
-
## Designing AI solutions with custom models
52
+
Help orchestrate data ingestion, labeling, evaluation, and iterative improvements at scale.
53
53
54
-
Solution architects should follow a structured, repeatable design approach to ensure models align with business objectives.
54
+
#### Agent and tooling integration
55
55
56
-
### Step 1 Define the business objectives
56
+
Custom models can be embedded into Foundry agents and orchestrations to support multi-step reasoning and automated workflows.
Map objectives to use cases where custom models outperform standard copilots.
60
+
Includes content filtering, safety evaluation, transparency artifacts, policy enforcement, and auditability.
61
61
62
-
###Step 2 Assess data requirements
62
+
#### Deployment topologies
63
63
64
-
Evaluate available proprietary datasets.
64
+
- Hosted secure environments.
65
+
- Private networking deployments.
66
+
- Integration with Azure Kubernetes Service and Foundry runtime environments.
65
67
66
-
Identify gaps in labeling, quality, diversity, or structure.
68
+
## Designing AI solutions with custom models
67
69
68
-
Ensure governance policies allow data to be used in model training.
70
+
Solution architects should follow a structured, repeatable design approach to ensure models align with business objectives.
69
71
70
-
### Step 3 Select the custom model path
72
+
### Step 1. Define the business objectives
71
73
72
-
#### Typical options include
74
+
- Identify measurable outcomes, such as accuracy goals, time-saved targets, and cost-efficiency goals.
75
+
- Map objectives to use cases where custom models outperform standard copilots.
73
76
74
-
**Fine-tuning foundation models**<br>Adjust behavior using domain datasets without full retraining.
77
+
### Step 2. Assess data requirements
75
78
76
-
**Training domain-built small models**<br>Useful for lightweight tasks requiring speed and edge compatibility.
79
+
- Evaluate available proprietary datasets.
80
+
- Identify gaps in labeling, quality, diversity, or structure.
81
+
- Ensure governance policies allow data to be used in model training.
77
82
78
-
**Hybrid architectures**<br>Combining custom models with prebuilt copilots for augmented reasoning.
83
+
### Step 3. Select the custom model path
79
84
80
-
### Step 4 Integration with enterprise systems
85
+
Typical options include:
81
86
82
-
#### Custom models should integrate with
87
+
#### Fine-tuning foundation models
83
88
84
-
Dynamics 365 applications
89
+
Adjust behavior using domain datasets without full retraining.
85
90
86
-
Azure Functions and Logic Apps
91
+
#### Training domain-built small models
87
92
88
-
Foundry agent workflows
93
+
Useful for lightweight tasks requiring speed and edge compatibility.
89
94
90
-
Azure AI Search and data stores
95
+
#### Hybrid architectures
91
96
92
-
Business process automation pipelines
97
+
Combine custom models with prebuilt copilots for augmented reasoning.
93
98
94
-
### Step 5 Validation and evaluation
99
+
### Step 4. Integrate with enterprise systems
95
100
96
-
#### Establish a rigorous testing plan
101
+
Custom models should integrate with:
97
102
98
-
Scenario-based evaluations
103
+
- Dynamics 365 applications.
104
+
- Azure Functions and Logic Apps.
105
+
- Foundry agent workflows.
106
+
- Azure AI Search and data stores.
107
+
- Business process automation pipelines.
99
108
100
-
Safety and bias analysis
109
+
### Step 5. Validate and evaluate
101
110
102
-
Stress, latency, and scaling tests
111
+
Establish a rigorous testing plan that includes:
103
112
104
-
ROI measurement and business validation
113
+
- Scenario-based evaluations.
114
+
- Safety and bias analysis.
115
+
- Stress, latency, and scaling tests.
116
+
- ROI measurement and business validation.
105
117
106
118
## Operationalizing custom models in Foundry
107
119
108
-
Modern AI systems require robust operational frameworks, especially with custom AI models. Foundry has tools to help support these custom AI models for longterm feasibility.
120
+
Modern AI systems require robust operational frameworks, especially for custom AI models. Foundry has tools to support these models for long-term feasibility.
109
121
110
122
### Key operational components
111
123
112
-
**Model Monitoring and Observability**<br>Track drift, performance degradation, user friction areas, latency, and unexpected model outputs.
124
+
#### Model monitoring and observability
113
125
114
-
**Governance and Compliance Controls**<br>Ensure every deployment meets enterprise privacy, security, and regulatory requirements.
126
+
Track drift, performance degradation, user friction areas, latency, and unexpected model outputs.
115
127
116
-
**Versioning and Lifecycle Management**<br>Maintain clear model version trails, update pipelines, and rollback strategies.
128
+
#### Governance and compliance controls
117
129
118
-
**Deployment Automation (MLOps/GenAIOps)**<br>Automate validations, approval workflows, and environment-specific deployments.
119
-
120
-
### Custom model decision matrix
130
+
Ensure every deployment meets enterprise privacy, security, and regulatory requirements.
121
131
122
-
Decision Factor | Standard Copilot | Custom Model (Foundry)
0 commit comments