AdRelevance Specification

Home / AdRelevance / AdRelevance Specification

AdRelevance

AdRelevance: Understanding Content, Not People.

AdRelevance is an open-source, privacy-first contextual advertising engine that uses multimodal AI to understand the semantic context of digital content and match advertisements to that context without tracking individual users.

The system analyzes text, images, and video as complementary sources of contextual information. Rather than determining what advertisements to display by building profiles of people, AdRelevance builds a semantic understanding of the content being viewed and uses that understanding to identify relevant advertising opportunities.

Overview

Traditional advertising systems frequently depend on behavioral tracking, personal profiling, cookies, and other mechanisms designed to understand individual users. AdRelevance takes a different approach by focusing its intelligence on the content itself.

The platform creates a semantic overlay across the content environment. Text, imagery, video scenes, audio transcripts, metadata, and other contextual signals can be analyzed independently and then combined into a unified representation of the content.

Advertisements are similarly described through semantic, visual, contextual, and campaign metadata. The matching system compares the content context against available advertising inventory and ranks potential placements according to relevance, compatibility, safety, and placement requirements.

The architecture is modular so that the core contextual intelligence can operate independently while optional plugins extend the platform with additional models, media processors, advertising exchanges, publisher integrations, analytics systems, and other capabilities.

Design Principles

Content Over People

AdRelevance is designed around contextual intelligence rather than behavioral surveillance. The primary object of analysis is the content environment, not an individual’s browsing history or behavioral profile.

Multimodal Understanding

Context should not be determined from text alone. Images and video can contain information that is absent from written content, while text can provide meaning that cannot be reliably inferred from visual information alone. AdRelevance therefore treats multiple modalities as complementary contextual signals.

Semantic Relevance

The system should evaluate meaning, relationships, intent, and contextual compatibility rather than relying solely on keyword matching.

Privacy First

The architecture should minimize unnecessary collection and processing of personal information. Contextual advertising should be possible without requiring persistent user profiles.

Modular Design

Core capabilities should remain independently testable and replaceable. Optional capabilities should be implemented through plugins rather than creating unnecessary dependencies in the core system.

Model Agnostic

AdRelevance should support interchangeable AI models and providers where practical. No single model, embedding provider, computer vision system, or inference platform should be required by the architecture.

Human Oversight

Advertising decisions should remain explainable and configurable. Publishers and administrators should be able to establish relevance thresholds, exclusions, brand safety policies, placement rules, and other controls.

Core Modules

Content Ingestion

The Content Ingestion module receives and normalizes contextual material from supported sources.

It should support:

Web content
Articles
Publisher pages
Text documents
Images
Video
Audio
Metadata
Structured content feeds

The module should identify available content modalities and prepare them for downstream analysis without requiring every source to provide the same format.

Text Context Analysis

The Text Context Analysis module evaluates written content and extracts semantic information.

It should identify:

Topics
Subjects
Entities
Concepts
Intent
Context
Relationships
Sentiment where appropriate
Content categories
Semantic embeddings

The module should distinguish between isolated keywords and the broader meaning of the surrounding content.

Image Context Analysis

The Image Context Analysis module evaluates images as independent contextual sources.

It should identify relevant:

Objects
Scenes
Activities
Environments
Visual concepts
Brands where appropriate
Products where appropriate
Text contained within images

Image analysis should contribute semantic information to the overall content context rather than functioning solely as object detection.

Video Context Analysis

The Video Context Analysis module analyzes video as a sequence of contextual events rather than treating a video as a single static asset.

It should support:

Scene segmentation
Frame analysis
Object recognition
Activity recognition
Visual context extraction
Audio analysis
Speech transcription
Temporal context
Scene-level semantic representations

The module should preserve relationships between events occurring at different points in a video.

Audio and Speech Analysis

The Audio and Speech Analysis module processes spoken or relevant audio content associated with media.

It should support:

Speech-to-text transcription
Speaker-independent semantic analysis
Topic extraction
Intent detection
Audio event classification where appropriate
Synchronization with video context

Audio-derived context should be treated as one signal among multiple modalities.

Semantic Overlay Engine

The Semantic Overlay Engine is the central contextual intelligence layer.

It combines signals generated by the text, image, video, audio, metadata, and other core analysis modules into a unified semantic representation.

The overlay should establish relationships between:

Textual concepts
Visual concepts
Video scenes
Audio concepts
Entities
Topics
User-visible content sections
Content intent
Advertising categories

The resulting semantic overlay represents the contextual environment in which an advertisement may be displayed.

Multimodal Context Fusion

The Multimodal Context Fusion module combines contextual representations from different modalities.

It should:

Normalize contextual signals
Weight modality-specific evidence
Resolve conflicting signals
Identify corroborating signals
Generate unified embeddings
Preserve modality-specific metadata
Support configurable fusion strategies

The system should not assume that every modality carries equal contextual value.

Advertisement Understanding

The Advertisement Understanding module creates semantic representations of available advertisements.

Advertisement records may include:

Advertiser
Campaign
Product or service
Topics
Categories
Intended audience context
Semantic embeddings
Visual characteristics
Brand requirements
Placement requirements
Geographic restrictions where applicable
Content exclusions
Campaign priorities
Availability windows

This allows advertisements to be evaluated semantically rather than through simple keyword matching.

Semantic Matching

The Semantic Matching module compares the semantic overlay of content with the semantic representation of available advertisements.

Matching should consider:

Topic relevance
Conceptual similarity
Contextual intent
Visual compatibility
Temporal relevance
Product relevance
Category compatibility
Advertiser requirements
Publisher requirements
Content exclusions
Brand safety requirements

The matching system should support configurable scoring strategies.

Ad Ranking

The Ad Ranking module orders eligible advertisements according to configurable relevance and policy criteria.

Ranking may incorporate:

Semantic similarity
Multimodal compatibility
Campaign priority
Placement suitability
Brand safety
Advertiser constraints
Publisher constraints
Availability
Performance signals where permitted
Quality thresholds

Ranking should remain explainable enough for publishers and administrators to understand why an advertisement was selected.

Placement Intelligence

The Placement Intelligence module determines where an advertisement can be displayed within the contextual environment.

It should evaluate:

Page sections
Content boundaries
Media boundaries
Available ad formats
Visual proximity
Content relevance
Placement suitability
Publisher-defined restrictions

The module should distinguish between selecting the right advertisement and selecting the right location for that advertisement.

Brand Safety and Content Policy

The Brand Safety and Content Policy module evaluates whether an advertisement is appropriate for a particular contextual environment.

It should support configurable policies for:

Sensitive subjects
Mature content
Violence
Tragedy
Illegal activity
Explicit material
Political content
Financial content
Medical content
Brand-specific exclusions
Publisher-specific exclusions

Policies should be configurable without requiring changes to the underlying AI models.

Privacy and Data Controls

The Privacy and Data Controls module establishes boundaries around contextual processing.

It should provide:

Data minimization
Configurable retention
Context-only processing
Removal of unnecessary identifiers
Processing controls
Privacy-preserving logs
Configurable data handling policies

The architecture should avoid requiring persistent individual user profiles for contextual ad placement.

Contextual Cache

The Contextual Cache module stores reusable contextual representations when permitted by configured privacy and retention policies.

It should support:

Context embeddings
Content fingerprints
Analysis results
Expiration policies
Cache invalidation
Modality-specific results

Caching should not become a mechanism for constructing persistent behavioral profiles.

Vector Search

The Vector Search module provides semantic retrieval for advertisements and contextual representations.

It should support:

Embedding storage
Similarity search
Filtering
Metadata queries
Hybrid retrieval
Configurable distance metrics
Pluggable vector databases

The module should remain independent from any single vector database implementation.

Ad Delivery

The Ad Delivery module provides ranked advertising results to supported publisher environments.

It should support:

Web placements
Native placements
Display advertisements
Video advertisements
Contextual overlays
API-based delivery
Publisher-defined placement formats

Delivery should remain separate from semantic analysis so that the contextual engine can be integrated with different publishing systems.

Explainability

The Explainability module provides contextual reasoning about ad selection.

It should be able to expose appropriate signals such as:

Matched topics
Related concepts
Relevant visual signals
Relevant video scenes
Matching categories
Placement reasoning
Policy decisions
Relevance scores

Explanations should avoid exposing sensitive internal data or unnecessary information about individual users.

Configuration and Policy Management

The Configuration and Policy Management module provides centralized configuration for the platform.

It should manage:

Matching thresholds
Ranking weights
Model selection
Content policies
Brand safety rules
Publisher restrictions
Ad categories
Retention policies
Plugin configuration
Performance settings

Configuration should be declarative where practical.

API and Integration Layer

The API and Integration Layer exposes AdRelevance capabilities to external applications.

It should support interfaces for:

Content analysis
Context retrieval
Advertisement registration
Advertisement matching
Ad ranking
Placement recommendations
Policy evaluation
System configuration
Plugin management

APIs should use stable contracts and versioning where appropriate.

Observability

The Observability module provides operational visibility without turning analytics into user surveillance.

It should support:

System metrics
Processing latency
Model performance
Matching quality
Error reporting
Resource utilization
Ad selection statistics
Context analysis statistics

Analytics should be designed around system and contextual performance rather than individual behavioral tracking.

Optional Plugin Modules

Optional plugins extend AdRelevance without requiring additional functionality in the core engine.

AI Model Plugins

AI Model plugins may provide alternative:

Language models
Embedding models
Vision models
Video models
Speech models
Classification models
Multimodal models

Plugins should follow defined interfaces so models can be replaced without redesigning the core architecture.

Vector Database Plugins

Vector database plugins may integrate external or local vector storage systems.

Possible implementations include:

Qdrant
Weaviate
Elasticsearch
Other compatible vector databases

Media Processing Plugins

Media Processing plugins may add support for additional:

Image formats
Video formats
Audio formats
OCR systems
Transcoding systems
Scene detection systems

Publisher Plugins

Publisher plugins may provide integrations for:

Content management systems
Websites
Blogs
News platforms
Video platforms
Publishing APIs
Ad placement frameworks

Advertising Network Plugins

Advertising Network plugins may connect AdRelevance to external advertising inventory or marketplaces.

These integrations should preserve the contextual matching architecture and should not require behavioral profiling.

Analytics Plugins

Analytics plugins may provide additional reporting and performance analysis.

Analytics implementations should respect the project’s privacy principles and should not introduce user surveillance into the core system.

Brand Safety Plugins

Brand Safety plugins may provide additional classification models, policy engines, or external safety services.

Geographic Context Plugins

Geographic Context plugins may optionally provide location-related contextual information where legally and technically appropriate.

Geographic functionality should not require persistent individual tracking.

Commerce Plugins

Commerce plugins may connect contextual signals with product catalogs, affiliate systems, or commerce databases.

These plugins can identify commercially relevant products or services associated with the analyzed content.

Content Management Plugins

Content Management plugins may connect AdRelevance to external content sources and provide automated contextual analysis for publisher environments.

Notification Plugins

Notification plugins may provide administrative notifications for:

Policy violations
System failures
Model failures
Campaign changes
Integration problems

Semantic Overlay

The semantic overlay is the defining feature of AdRelevance.

Rather than treating a page as a collection of isolated keywords, the overlay creates a contextual representation across the complete content environment.

For a page containing an article, photograph, and embedded video, the system may identify:

The article’s primary subject
Secondary topics
Entities referenced in the text
Objects visible in the photograph
Activities occurring in the video
Concepts discussed in the video’s audio
Relationships between these signals
Appropriate advertising categories

The combined representation can then be compared against advertising inventory.

This approach allows an advertisement to be relevant even when the exact words associated with the advertisement do not appear in the content.

Contextual Relevance Scoring

AdRelevance should support a configurable relevance model that evaluates multiple signals.

Potential scoring dimensions include:

Semantic similarity
Topic alignment
Intent alignment
Visual compatibility
Video compatibility
Entity relationships
Brand safety
Placement suitability
Campaign requirements
Publisher requirements

No single scoring strategy should be mandatory. Implementations should be able to experiment with different ranking approaches while preserving the same core interfaces.

Advertising Context Without Behavioral Profiling

AdRelevance is designed to demonstrate that useful advertising does not require an individual behavioral profile.

The platform should prioritize contextual signals such as:

What the content is about
What is being shown
What activity is taking place
What concepts are being discussed
What products or services are relevant
Where an advertisement would naturally fit

The system should not require knowledge of what an individual previously viewed on unrelated sites to determine contextual relevance.

Performance

AdRelevance should be designed for low-latency contextual decision making.

Performance goals may include:

Efficient multimodal processing
Cached contextual representations
Parallel modality analysis
Fast semantic retrieval
Configurable model execution
Asynchronous processing where appropriate
Low-latency ad ranking
Scalable inference

A target response time of under 150 milliseconds may be pursued for suitable real-time placement workflows, recognizing that full multimodal analysis may require asynchronous processing depending on content size and model complexity.

Security

Security should be treated as a core architectural concern.

The system should provide:

Secure API authentication
Authorization controls
Input validation
Safe media processing
Plugin isolation where practical
Secure model loading
Dependency management
Configuration protection
Audit capabilities
Secure secrets handling

External content should never be treated as trusted input.

Extensibility

AdRelevance should allow developers to replace or extend individual components without modifying unrelated parts of the platform.

Core interfaces should support:

Alternative AI models
Alternative embedding systems
Alternative vector databases
Alternative media processors
Alternative ad delivery systems
Alternative policy engines
Alternative publisher integrations

Plugins should be independently configurable and should not become mandatory dependencies for core contextual functionality.

Developer Experience

The project should provide clear documentation for:

Installation
Configuration
API usage
Model integration
Plugin development
Publisher integration
Advertisement registration
Semantic matching
Policy configuration
Testing
Deployment
Contribution

Examples should demonstrate how to build integrations without requiring proprietary infrastructure.

Testing

The project should include testing for:

Content ingestion
Text analysis
Image analysis
Video analysis
Multimodal fusion
Semantic matching
Ad ranking
Placement selection
Brand safety
Privacy controls
API behavior
Plugin interfaces
Security boundaries
Performance

Multimodal test cases should evaluate whether combining multiple signals produces better contextual decisions than relying on a single modality.

Project Goals

AdRelevance aims to provide:

A privacy-first alternative to behavioral advertising
Multimodal contextual intelligence
Semantic advertising relevance
Explainable ad placement
Modular AI infrastructure
Open-source advertising infrastructure
Publisher-controlled contextual advertising
Extensible advertising integrations
Reduced dependence on user surveillance

The broader goal is to demonstrate that advertising can become more relevant by understanding content rather than people.

Contributing

Contributions are welcome from developers, researchers, publishers, advertisers, designers, privacy advocates, and other contributors interested in building better contextual advertising infrastructure.

Potential contributions include:

AI models
Semantic matching improvements
Multimodal analysis
Ranking algorithms
Publisher integrations
Advertising integrations
Plugin modules
Privacy improvements
Security improvements
Documentation
Testing
Performance optimization

Please review the project’s contribution guidelines and notice requirements before submitting changes.

Vision

Advertising should be able to understand context without turning people into behavioral profiles.

AdRelevance builds the infrastructure for a different model: multimodal AI that understands the content, identifies meaningful relationships, and places advertising where it is genuinely relevant.

AdRelevance: Understanding Content, Not People.

Specification Branding License (SBL)

Standard

Fully AGPL-3.0+ compliant system
Copyleft enforced for network deployments
Required attribution:
- Roxanne Ardary
- roxanneardary.com

Optional

Specification Branding License (SBL)
- Attribution-free commercial deployment
- Pricing based on scale, usage, and deployment scope
- https://roxanneardary.com/adrelevance/

License & Notice Requirements

AdRelevance is released under the GNU Affero General Public License v3.0 or later (AGPL-3.0+).
By contributing to this project, you agree that your contributions will also be released under this license.

Please note the following:

All contributions must comply with the AGPL-3.0+ terms.
Under Section 7 of the license, all redistributions, forks, and derivative works must preserve attribution to:
Roxanne Ardary and roxanneardary.com.
AdRelevance specifications are free to use with attribution. A Specification Branding License can be negotiated upon request.
The project’s notice.md file tracks attribution requirements and contributor acknowledgments. Any update that adds new contributors or modifies attribution should also update notice.md.
When submitting a pull request, ensure that any new files maintain the attribution headers where applicable.
Network-deployed versions of this software must also remain fully AGPL-3.0+ compliant, including exposure of source code modifications when applicable under the license.

For full legal details, please refer to the AGPL-3.0+ license and the project’s notice.md file.

Notice – AdRelevance

Attribution Requirement:
Under Section 7 of the AGPL 3.0+ license, all redistributions, forks, and derivative works, including network-deployed versions of this project, must provide attribution to Roxanne Ardary and roxanneardary.com.

Contributors

This file tracks contributors and their specific contributions to the project.

Roxanne Ardary, roxanneardary.com – March 9, 2026
Created the repository for AdRelevance. Developed the open-source privacy-first contextual advertising engine using semantic overlays and multimodal AI for text, image, and video content.
[Add other contributors here] – [Date]
[Describe contribution in one sentence]

License – AdRelevance

This repository is licensed under the GNU Affero General Public License v3.0 or later (AGPL-3.0+).

Key Points:

You are free to use, modify, and distribute the code.
All redistributions, forks, and derivative works or network-deployed versions must also be licensed under AGPL-3.0+ and provide attribution to Roxanne Ardary and roxanneardary.com as required under Section 7 of the license.
The software is provided “as is,” without warranty of any kind.

For the full license text, see GNU AGPL-3.0 License.