Sectors & YAML

Every Queria company belongs to a sector (sector) and optionally to a sub-sector (subSector). The configuration lives in versioned YAML files in the repository, describing the semantic domain of the sector: which entities exist, how they are called in natural language, which measures to aggregate, which access rules to apply.

From v3.5.0 these YAMLs are much richer: they include tags, domain terms, dimensions with aliases, measures, hierarchies and confidentiality policies.

Available sectors

Queria includes 26+ out-of-the-box sectors:

Category	Sectors
Professional services	LEGAL, FINANCE, INSURANCE, BANKING, HR
Healthcare	HEALTHCARE, PHARMA, RESEARCH
Industry	MANUFACTURING, AUTOMOTIVE, ENERGY, ENGINEERING, CONSTRUCTION
Consumer goods	FOOD, ECOMMERCE
Transport	LOGISTICS, TELECOM
Public	GOVERNMENT, EDUCATION, NONPROFIT
Services	HOSPITALITY, TOURISM, SUPPORT, MEDIA, REAL_ESTATE, AGRICULTURE
Generic	CONTRACT, TECHNICAL, GENERIC, CUSTOM

Each sector has a YAML file in backend/src/config/sectors/<sector>.sector.yml and is loaded at startup by the sector-loader.

Anatomy of a sector YAML

Example (excerpt from legal.sector.yml):

sector: LEGAL
name: "Legale"
description: "Contracts, judgments, regulations, legal acts"

typicalConcepts:
  - tipo_documento
  - parti
  - giurisdizione
  - data
  - materia
  - articolo
  - clausola

typicalQueries:
  - "Find contracts with company X"
  - "Which judgments relate to condominiums?"
  - "What does article 5 of the contract say?"
  - "Privacy regulation"

relevantMetadataFields:
  - document_type
  - parties
  - jurisdiction
  - date
  - matter
  - court

externalSources:
  - legal-sources

domainTerms:
  - contratto
  - sentenza
  - clausola
  - articolo
  - giurisdizione
  - parte
  - attore
  - convenuto

dimensions:
  - name: document_type
    type: string
    aliases: ["tipo documento", "tipo atto", "tipologia"]
    hierarchy: [category, specific_type]
  - name: parties
    type: string[]
    aliases: ["parti", "soggetti", "attore", "convenuto"]
  - name: date
    type: date
    aliases: ["data", "anno", "periodo"]

measures:
  - name: count
    type: count
    description: "Element count"
  - name: distribution
    type: group_by
    description: "Distribution by dimension"

accessPolicy:
  topicIsolation: true
  confidentialityRules:
    - role: READER
      exclude: ["CONFIDENTIAL", "RESTRICTED"]
    - role: EDITOR
      exclude: ["RESTRICTED"]
    - role: ADMIN
      exclude: []

yamlVersion: 1

Sector fields

Identity

Field	Type	Description
sector	enum	Unique code (uppercase, e.g. LEGAL)
name	string	Display name (e.g. "Legale")
description	string	Short description shown in the company UI

Natural language

Field	Type	Use
typicalConcepts[]	string[]	Concepts that recur in questions, used by the planner
typicalQueries[]	string[]	Typical questions, used as classifier prompt
domainTerms[]	string[]	Sector vocabulary: boosts chunks containing them and improves metadata extraction

Tags and domain terms boosted in v3.5.0

domainTerms are now also stamped on Qdrant chunks as a top-level payload (domainTerms) via the sector-fields backfill. This speeds retrieval filters ~5x and improves the relevance booster.

Dimensions and measures

dimensions[] -- semantic attributes worth filtering, grouping, comparing on:

• name: canonical name (e.g. parties)
• type: string | string[] | number | date
• aliases[]: natural-language synonyms (e.g. ["parti", "soggetti", "attore", "convenuto"])
• hierarchy[]: optional hierarchies (e.g. [category, specific_type] for drill-down)

measures[] -- applicable aggregations:

• type: count | sum | avg | min | max | group_by

The planner LLM uses dimensions and measures to build structured queries when the user asks for aggregations ("how many contracts in 2025?", "average amount per category?").

External sources

externalSources[] -- lists external microservices relevant for the sector. Examples:

• LEGAL -> legal-sources (Normattiva)
• FOOD -> food-sources (Open Food Facts)
• PHARMA -> pharma-sources (FDA, ClinicalTrials)
• TOURISM -> consumes open-data via AI Constructor

Access policy

Role-based confidentiality policies:

accessPolicy:
  topicIsolation: true
  confidentialityRules:
    - role: READER
      exclude: ["CONFIDENTIAL", "RESTRICTED"]

Meaning: a user with role READER cannot see chunks with confidentiality in CONFIDENTIAL or RESTRICTED. Policies apply to Qdrant filtering at runtime, independently of topics.

Sub-sectors

A sector can have sub-sectors (subSectors) that inherit the base and apply deltas:

subSectors:
  contractDrafting:
    description: "Contract drafting and review"
    dimensions:
      - name: clause_category
        type: string
        aliases: ["clause category", "clause type"]

Sub-sectors are useful for companies that share the base sector but have specializations: e.g. one "tax" and one "labor" law firm both have LEGAL sector but different sub-sectors with finer dimensions.

In DB every sub-sector is a SubSector row; assignment to the company is via Company.subSectorId. Without assignment, sector routing does not kick in and Qdrant chunks are not enriched with sector fields.

Assigning a sector to a company

From the admin panel:

1. Companies > [company] > Sector -> select the sector from the dropdown.
2. Optional: choose a sub-sector.
3. Save. From now on, newly ingested documents will be stamped with sector fields.

For pre-v3.5.0 companies:

UPDATE "Company"
SET "subSectorId" = '<id-from-SubSector>'
WHERE id = '<companyId>';

Then run backfills to enrich existing chunks (see Ingestion DSL > Reingest).

Editing a YAML

*.sector.yml files are repo-versioned. To edit them:

1. Open a PR on backend/src/config/sectors/<sector>.sector.yml.
2. Schema validated by Zod (_schema.ts): CI build fails if invalid.
3. Bump yamlVersion by 1.
4. Deploy: at boot, sector-loader reloads YAMLs, sector-seeder updates Sector and SubSector in DB.

Backfill after edit

If you add/rename domainTerms, consider rerunning the sector-fields backfill on the sector's companies, otherwise existing chunks won't have the new terms in payload.

Custom sector per tenant

For tenants that don't fit the 26 bundled sectors:

• Use sector: CUSTOM with an internal YAML.
• Configure dimensions/measures/domain terms specific to the customer domain.
• Documents are still indexed and searched normally.

UI-based custom sector creation is on the V2 roadmap (no more YAML PRs).

Company tags (boosted in v3.5.0)

Regardless of sector, every company has a tags[] field that works as free labels:

• Visible in the company sheet.
• Filterable in cross-tenant searches (SYSTEM_ADMIN only).
• Used by the planner to disambiguate behavior when sector alone isn't enough (e.g. "company tagged enterprise" -> use enterprise bot template).

Add tags in the Company > Tags sheet, separated by enter or tab.

Company domain terms

Beyond the sector domainTerms, every company can add company-specific domain terms from its sheet:

• Internal acronyms (e.g. MOL, EBITDA-adj).
• Customer product names.
• Company terminology glossary.

These are merged with the sector ones when evaluating chunk relevance, giving an explicit boost to the company lexicon.

---

Queria v3.5.0 -- Boosted Sector YAMLs