selectry taxonomy MVP implementation runbook

This runbook implements the portal-first Azure AI Search ESCO/O*NET skill matching MVP.

1. Azure Storage

Create a Storage Account:

• Resource group: rg-powerplatform-billing
• Name: selectrytaxonomydata
• Region: Germany West Central
• Performance: Standard
• Redundancy: LRS

Create a private Blob container:

• Container name: taxonomy-source

2. Azure AI Search index

In azureaisearchselectry, create index selectry-skills-v2.

Use the JSON schema in:

taxonomy_mvp/selectry-skills-v2.index.json

3. selectry-local seed data

Upload modern tool data directly to the new index.

Request:

POST https://azureaisearchselectry.search.windows.net/indexes/selectry-skills-v2/docs/index?api-version=2024-07-01

Headers:

Content-Type: application/json
api-key: <admin-key>

Body:

taxonomy_mvp/selectry-local-skills.json

4. ESCO transform

Download English ESCO skills/competences CSV from:

https://esco.ec.europa.eu/en/use-esco/download

Transform it:

python3 taxonomy_mvp/prepare_taxonomy.py \
  --source ESCO \
  --input /path/to/esco_skills_en.csv \
  --output taxonomy_mvp/out/esco-skills-upload.json

Upload taxonomy_mvp/out/esco-skills-upload.json to the same Azure Search index using the same /docs/index endpoint as above.

5. O*NET transform

Download O*NET database files from:

https://www.onetcenter.org/

Start with skills, knowledge, abilities, and technology skills files.

Transform each file:

python3 taxonomy_mvp/prepare_taxonomy.py \
  --source ONET \
  --input /path/to/onet_file.csv \
  --output taxonomy_mvp/out/onet-upload.json

Upload each transformed JSON file to selectry-skills-v2.

6. Search Explorer tests

Use selectry-skills-v2.

Test software development:

{
  "search": "software development",
  "queryType": "simple",
  "top": 5,
  "select": "id,source,preferredLabel,altLabels,description,skillType,uri"
}

Test Docker:

{
  "search": "Docker",
  "queryType": "simple",
  "top": 5,
  "select": "id,source,preferredLabel,altLabels,description,skillType,uri"
}

Test project management:

{
  "search": "project management",
  "queryType": "simple",
  "top": 5,
  "select": "id,source,preferredLabel,altLabels,description,skillType,uri"
}

7. Power Automate update

In PA_GetSkillTaxonomyMatches_v2, update the HTTP URI:

https://azureaisearchselectry.search.windows.net/indexes/selectry-skills-v2/docs/search?api-version=2024-07-01

Keep the body expression:

json(concat('{"search":"', item(), '","queryType":"simple","top":3,"select":"id,source,preferredLabel,altLabels,description,skillType,uri"}'))

The last action must be Return value(s) to Power Virtual Agents, not generic Response.

Return:

taxonomyStatus = taxonomy_lookup_completed
taxonomyMatchesJson = string(variables('varResults'))
taxonomySourcesUsedJson = ["ESCO","O*NET","selectry-local"]

8. Copilot Studio tool and agent

Tool display name:

Match job skills against ESCO and O*NET skill databases

Underlying flow:

PA_GetSkillTaxonomyMatches_v2

Model:

Claude Sonnet 4.6

Behavior:

• Input is role core JSON from Agent 1.
• Extract unique skill names.
• Call the taxonomy flow with a JSON string array.
• Preserve original priority and source.
• Use taxonomy only as enrichment.
• Do not invent ESCO/O*NET matches.
• Do not promote optional/unconfirmed signals to must-have.