Brand Brain – Product Build for Data & Platform Engineers
1. What is the Brand Brain?
The Brand Brain is a centralised repository of machine-readable brand knowledge that AI tools can query in real time.
It stores, connects, and serves:
- Brand rules & policies (mustDo / mustNotDo)
- Voice & tone guidance (verbal identity)
- Visual & audio identity metadata + asset pointers
- Context-specific overrides (by region, channel, archetype, segment)
- Canonical good/bad examples for few-shot prompting
It exposes this via APIs so downstream tools (copy generators, design aids, personalisation engines) can reliably “load the brand brain” before calling an LLM.
2. Primary Use Cases
- LLM-powered content tools
- Marketing copy (CRM, social, web, ads)
- UX & support responses
- On-site merchandising / personalisation
- Interactive, searchable repository
- Brand, creative, CX and legal can browse, edit, and approve:
- Rules, tone, tokens, examples
- Governance & evolution
- Central place to update rules and style once and have all tools pick it up
- Future: feed performance & feedback back into rules
3. Scope (v1)
In scope:
- Graph representation of:
Brand,Context,CustomerSegment,ArchetypeVerbalIdentityToken,VisualIdentityToken,AudioIdentityTokenContentGuideline(mustDo / mustNotDo / advisory)BrandExample(good/bad examples for LLMs)- References (URIs) to binaries:
- Logos, palettes, templates, audio samples
- API for:
- Fetching a Brand+Context snapshot for LLM usage
- Searching rules/examples/tokens
- Basic UI for:
- Browsing a brand’s brain
- Editing rules & tokens
- Managing examples
Out of scope for v1 (but designed for later):
- Performance metrics & optimisation loop
- Automatic rule updates from metrics
- Full multi-brand, multi-market analytics
4. High-Level Architecture
Three main layers:
- Graph Layer (source of truth)
- Property graph DB (e.g. Neo4j / Neptune / JanusGraph)
- Stores entities & relationships:
- Brands, contexts, segments, archetypes
- Identity tokens (verbal, visual, audio)
- Content guidelines
- Brand examples
- Pointers to binary assets
-
This is where governance and linking happens.
-
Blob / Asset Layer
- Object storage / DAM (S3/GCS/etc.)
- Stores binary assets:
- Logos, brand decks, PDFs, creative, audio files
-
Graph nodes hold URIs + metadata.
-
Search & Vector Layer
- Full-text index over:
ContentGuideline,VerbalIdentityToken,BrandExample,Brand
- Vector index (optional v1.5+):
- Embeddings for examples / style docs
- Backed by OpenSearch/ES, pgvector, or a vector DB.
5. Core Data Model Slice
This is the Brand Brain slice of the wider Brand KG.
Key Node Types
Brand- Master brand / owned / hosted / co-brand.
-
Fields:
id,name,brand_type,description,tagline_default,primary_language,localized_*,metadata. -
Context - Composite of brand + channel + region + segment/archetype.
- Fields:
id,name,description,metadata. -
Linked to:
Brand,GeographicRegion,CustomerSegment,Archetype. -
CustomerSegment - Operational segment from CDP.
-
Fields:
id,name,criteria,metadata. -
Archetype - Synthetic or curated archetype (persona).
- Fields:
id,name,description,archetype_type,generation_method,feature_profile,language_codes,is_active,metadata. -
Linked to segments, trends, need states, and (optionally) real customers.
-
VerbalIdentityToken - Machine-readable voice & tone.
-
Fields:
id,name,tone_of_voice,style_guidelines,required_elements,approved_keywords,prohibited_keywords,example_snippets,metadata. -
VisualIdentityToken - Visual system metadata.
-
Fields:
id,name,primary_colors,secondary_colors,logo_refs,imagery_style,iconography,metadata. -
AudioIdentityToken - Audio brand cues.
-
Fields:
id,name,audio_logo_ref,music_style,voice_guidelines,metadata. -
ContentGuideline - Rules & policies.
-
Fields:
id,rule_type(MUST_DO,MUST_NOT_DO,ADVISORY),name,description,dimension,severity,source,metadata. -
BrandExample - Canonical examples for few-shot prompts.
-
Fields:
id,name,description,example_type(GOOD,BAD,REFERENCE),modality(TEXT,IMAGE,AUDIO, …),channel,language,body_text,asset_refs,labels,created_by,metadata. -
Asset(optional, generic media) - Fields:
id,asset_type,uri,mime_type,metadata.
Key Relationships (Simplified)
- Brand to identity:
(Brand)-[:USES_VERBAL_IDENTITY]->(VerbalIdentityToken)(Brand)-[:USES_VISUAL_IDENTITY]->(VisualIdentityToken)-
(Brand)-[:USES_AUDIO_IDENTITY]->(AudioIdentityToken) -
Brand + context:
(Context)-[:CONTEXT_FOR_BRAND]->(Brand)(Context)-[:CONTEXT_FOR_REGION]->(GeographicRegion)(Context)-[:CONTEXT_FOR_SEGMENT]->(CustomerSegment)-
(Context)-[:CONTEXT_FOR_ARCHETYPE]->(Archetype) -
Rules:
(Brand)-[:HAS_CONTENT_GUIDELINE]->(ContentGuideline)(InternalCategory/GpcCategory)-[:HAS_CONTENT_GUIDELINE]->(ContentGuideline)-
(ContentGuideline)-[:OVERRIDES_IN_CONTEXT]->(Context) -
Examples:
(Brand)-[:HAS_EXAMPLE]->(BrandExample)(Context)-[:USES_EXAMPLE]->(BrandExample)(ContentGuideline)-[:ILLUSTRATED_BY]->(BrandExample)(BrandExample)-[:TARGETS_ARCHETYPE]->(Archetype)-
(BrandExample)-[:TARGETS_SEGMENT]->(CustomerSegment) -
Assets:
(VisualIdentityToken)-[:USES_ASSET]->(Asset)(AudioIdentityToken)-[:USES_ASSET]->(Asset)(BrandExample)-[:USES_ASSET]->(Asset)
This is for Brand Brain v1.
6. Brand Brain APIs (v1)
6.1 Context Snapshot for LLMs
Goal: Return everything an LLM needs for a given brand+context in one call.
Endpoint (conceptual):
GET /brand-brain/context
Query params:
brand_id(required)region_code(optional)channel(optional)language(optional)archetype_idorsegment_id(optional)include_examples(default: true)max_examples(default: 10, configurable)
Response (shape):
{
"brand": { ... },
"context": { ... },
"verbal_identity": { ... },
"visual_identity": { ... },
"audio_identity": { ... },
"guidelines": {
"must_do": [ ... ],
"must_not_do": [ ... ],
"advisory": [ ... ]
},
"examples": [ ... ],
"risk_domains": [ ... ],
"cultural_signals": [ ... ]
}
6.2 Search
Goal: Power UI + tooling search across rules, tokens, examples.
Endpoint:
GET /brand-brain/search
Params:
q– free text- Filters:
brand_id,entity_types,region_code,channel,language,archetype_id - Pagination:
limit,offset
6.3 Admin / CRUD (Brand Teams)
POST /brands/{brand_id}/guidelinesPATCH /guidelines/{guideline_id}POST /brands/{brand_id}/verbal-identityPATCH /verbal-identity/{token_id}POST /brands/{brand_id}/examplesPATCH /examples/{example_id}
All write operations:
- Update graph
- Reindex search/vector layer
7. UI Expectations (for Context)
Engineers don’t need to build all of this in v1, but the data model should support:
-
Brand overview page:
-
Summary, positioning, identity tokens, key rules
-
Context explorer:
-
Filters: brand, region, channel, language, archetype
- Shows the resolved rules, tokens, and examples for that slice
-
Search:
-
Keyword search over all rules/examples/tokens
-
Example gallery:
-
Good/bad examples with linked rules and contexts
-
Change history:
-
Basic audit log on rules & tokens
8. Implementation Notes & Priorities
v1 priorities:
-
Stand up graph layer with the Brand Brain slice:
-
Brand,Context,CustomerSegment,Archetype - Identity tokens, guidelines, examples
- Implement Context Snapshot API (
/brand-brain/context). - Implement Search API over rules/examples/tokens.
- Wire up basic admin CRUD for brand teams.
- Ensure config & deployment pattern so tools can call Brand Brain synchronously before LLM use.
Key technical considerations:
-
Indexes:
-
Unique constraints on
idfields. - Index on
Brand.name,Brand.brand_type. -
Full-text index for
ContentGuideline.description,VerbalIdentityToken.style_guidelines,BrandExample.body_text. -
Versioning:
-
Prefer simple versioning on rules/tokens:
- Properties:
version,valid_from,valid_to. - Alternatively, create successor nodes with
PREVIOUS_VERSION_OF.
- Properties:
-
Security:
-
No PII in Brand Brain.
-
Access control for write APIs:
- Brand / market / role-based permissions.