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
Copy file name to clipboardExpand all lines: learn-pr/wwl/design-alm-process-ai-powered-business-solutions/includes/2-design-alm-process-data-used-ai-models-agents.md
+36-35Lines changed: 36 additions & 35 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,28 +1,28 @@
1
1
## Overview
2
2
3
-
This expert unit equips solution architects to design an application lifecycle management (ALM) process for **all data that powers AI models and agents**. You'll define data-centric stages, controls, roles, and promotion gates that keep datasets, prompts, knowledge sources, and telemetry **governed, reproducible, and compliant** from inception through retirement. The approach integrates enterprise data governance, regional data movement considerations, and operational monitoring so that AI solutions scale safely across environments.
3
+
This expert unit equips solution architects to design an application lifecycle management (ALM) process for **all data that powers AI models and agents**. You define data-centric stages, controls, roles, and promotion gates that keep datasets, prompts, knowledge sources, and telemetry **governed, reproducible, and compliant** from inception through retirement. The approach integrates enterprise data governance, regional data movement considerations, and operational monitoring so that AI solutions scale safely across environments.
4
4
5
5
## Learning objectives
6
6
7
-
By the end of this unit, learners will be able to:
7
+
By the end of this unit, learners are be able to:
8
8
9
-
Map the **AI data lifecycle** to ALM stages and define artifacts, owners, and promotion gates.
9
+
*Map the **AI data lifecycle** to ALM stages and define artifacts, owners, and promotion gates.
10
10
11
-
Design **environment strategies** (Dev/Test/PreProd/Prod) that isolate data and enforce guardrails.
11
+
*Design **environment strategies** (Dev/Test/PreProd/Prod) that isolate data and enforce guardrails.
12
12
13
-
Specify **data residency and movement** policies for Copilot and agent features.
13
+
*Specify **data residency and movement** policies for Copilot and agent features.
14
14
15
-
Operationalize **quality, lineage, and drift** checks with go/nogo criteria and rollback plans.
15
+
*Operationalize **quality, lineage, and drift** checks with go/nogo criteria and rollback plans.
16
16
17
-
Establish a **RACI** and change workflows for datasets, knowledge sources, and evaluation telemetry.
17
+
*Establish a **RACI** and change workflows for datasets, knowledge sources, and evaluation telemetry.
18
18
19
19
## ALM foundations for AI data
20
20
21
21
### Scope the "data" in AI ALM
22
22
23
23
#### Treat the following as versioned, promotable ALM artifacts:
* Execute **canary runs** using masked/representative Prodlike data.
85
+
* Execute **canary runs** using masked/representative Prod like data.
86
86
87
-
*_Canary testing is a low-risk deployment strategy that acts as a form of "production regression testing" by releasing new code to a small, isolated subset of users or servers to identify issues before a full rollout._
87
+
*_Canary testing is a low-risk deployment strategy.
88
+
It releases new code to a small, isolated subset of users or servers to identify issues before a full rollout._
88
89
89
90
* Freeze **gold datasets** and sign **immutability attestations**.
90
91
91
92
***Gate D → E:** CAB approval; deployment runbooks and rollback plans ready.
92
93
93
-
### Phase E — Deploy & Serve
94
+
### Phase E—Deploy & Serve
94
95
95
96
* Promote gold corpora and indexes; **semantic indexing** or retrieval stores refreshed.
96
97
97
98
* Enforce **region/residency** settings and connector allow/deny lists.
98
99
99
100
* Register release in **data catalog** and publish **consumer contracts**.
* Run scheduled **reevaluation** with golden sets; file backlog items with trace IDs.
108
+
* Run scheduled **re-evaluation** with golden sets; file backlog items with trace IDs.
108
109
109
-
### Phase G — Evolve & Retire
110
+
### Phase G—Evolve & Retire
110
111
111
112
* Rotate or retrain on updated gold sets; **retest** before promote.
112
113
@@ -116,31 +117,31 @@ Establish a **RACI** and change workflows for datasets, knowledge sources, and e
116
117
117
118
## Controls & checks (what to verify at each promotion gate)
118
119
119
-
### Text table — ALM gate checks
120
+
### Text table—ALM gate checks
120
121
121
122
|**Gate**|**Control area**|**Questions to answer**|**Evidence required**|
122
123
|---|---|---|---|
123
124
|**A→B**| Catalog & ownership | Is the owner accountable? Is sensitivity labeled? | Data contract; catalog record; label policy proof |
124
-
|**B→C**| Data quality & lineage | Is data profiled, balanced, deidentified where needed? | Profiling report; lineage graph; version IDs |
125
-
|**C→D**| Evaluation & safety | Do evals meet thresholds? Any bias or unsafe patterns? | Metrics pack; safety report; model/data card |
125
+
|**B→C**| Data quality & lineage | Is data profiled, balanced, de-identified where needed? | Profiling report; lineage graph; version IDs |
126
+
|**C→D**| Evaluation & safety | Do evaluations meet thresholds? Any bias or unsafe patterns? | Metrics pack; safety report; model/data card |
126
127
|**D→E**| Compliance & residency | Do region rules and DLP policies permit use? | Residency mapping; DLP rules; approval memo |
127
128
|**E→F**| Runtime readiness | Can we monitor, roll back, and cap costs? | Dashboards; alarms; rollback plan; budget guard |
128
129
129
-
## Region, residency, and crossborder movement
130
+
## Region, residency, and cross-border movement
130
131
131
-
Document **where** prompts/outputs may be processed for Copilot and Power Platform features, and when **crossregion capacity** is required.
132
+
Document **where** prompts/outputs may be processed for Copilot and Power Platform features, and when **cross region capacity** is required.
132
133
133
134
In regulated scenarios, set the default to **inregion** and require explicit approval to enable **overflow processing**.
134
135
135
136
Align **mailbox region** (for activity data) and environment geo with your policy; define exceptions and purge schedules.
136
137
137
-
Professional visual — Residency decision tree (text)<br>Inregion capacity available? → Yes: keep local.<br>No: Is overflow allowed for this workload tier? → If yes, enable crossregion under admin control; else block feature or defer.
138
+
Professional visual—Residency decision tree (text)<br>Inregion capacity available? → Yes: keep local.<br>No: Is overflow allowed for this workload tier? → If yes, enable cross region under admin control; else block feature or defer.
138
139
139
140
## Roles and RACI
140
141
141
-
## Text RACI — data used in AI models and agents
142
+
## Text RACI—data used in AI models and agents
142
143
143
-
The below RACI chart is a representative sample for an organization. It is up to the architect on the project to validate and adjust the roles and responsibilities as appropriate for the implementation.
144
+
The below RACI chart is a representative sample for an organization. It's up to the architect on the project to validate and adjust the roles and responsibilities as appropriate for the implementation.
0 commit comments