Building an AI-Powered Advertisement Recommendation Engine for an OTT Platform
Overview
OTT platforms do not struggle with content volume.
They struggle with relevance.
Thousands of videos compete for attention, while ads often appear without context, timing, or intent. Viewers disengage. Advertisers miss impact. Platforms lose value.
This project focused on building the intelligence layer behind an OTT platform that understands video context, audience behavior, and precise ad placement.
The result was an AI-driven content discovery and recommendation system that analyzes video scenes, profiles users, and matches advertising to the right moment, screen, and audience.
Country
Italy
Technology
AI/ML · Computer Vision · Recommendation Systems · Cloud Infrastructure
Industry
OTT · Media · Advertising Technology
The Challenge: Rich Video, Poor Context
Lack of Segment-level Video Intelligence
OTT content was treated as a single asset, lacking visibility into scenes, moods, visuals, audio cues, and contextual shifts within the video.
Limited Personalization Across Devices
User targeting relied on broad preferences, offering minimal personalization across screens, devices, and viewing situations.
Context-Blind Advertising Placement
Ads were delivered without understanding video sentiment or timing, leading to irrelevant placements that reduced viewer engagement and advertiser impact.
Broad Targeting, Missed Revenue Opportunities
Advertisers were forced into generic campaigns, missing precise moments, niche audiences, and hyper-local targeting opportunities tied to specific video contexts.
What Was Built: An AI-Powered Content and Advertising Intelligence System
Kreeda Labs built the intelligence backbone of the platform by developing the Content Platform and the Advertising Platform.
AI-Based Content Discovery System
At the core is an AI system that analyzes video scene by scene.
It identifies and classifies:
- Sentiment such as cheerful, gloomy, energetic, or sentimental
- Context including sports, dialogue, fantasy, action, or drama
- Places like cities, mountains, sea, interiors, hotels, or homes
- Dominant colors and visual tone
- Sound characteristics including dialogue, soundtrack type, and ambient audio
- Objects, locations, and people within scenes
Each video becomes a structured intelligence asset rather than just media.
Advertising Management and Recommendation Engine
The advertising system connects three critical data layers.
Cataloged and Categorized Video Content
All videos enriched with scene-level intelligence generated by the content discovery system.
Profiled Users
User profiles built using:
- Explicit inputs such as registration data and preferences
- Implicit signals such as viewing behavior, interactions, searches, and ad engagement
Advertising Campaigns
Campaigns created by partners, media agencies, and internal teams, each with defined goals, targeting logic, and delivery rules.
Continuously Evolving Scene Categorization
The system adapts over time.
- Learns from advertiser behavior and usage patterns
- Refines semantics and linguistic understanding
- Improves categorization logic as interactions increase
This ensures the intelligence layer stays relevant as content and advertising strategies evolve.
Intelligent Matching Across Content, Users, and Ads
By intersecting these data layers, the system can:
- Deliver the right ad to the right user
- At the right moment in the content
- On the most relevant device
This enables both large-scale campaigns and highly targeted, context-aware advertising tied to specific scenes or moods.
Project Goals
Go beyond basic metadata
Break videos into segments and extract meaningful attributes.
Improve advertising effectiveness
Match campaigns to the right user, moment, and device.
Connect content, users, and advertising
Build a system where each layer informs the other.
Build a system that learns over time
Continuously refine intelligence based on real usage.
Technology Stack
Tech Stack | Tools |
|---|---|
AI and Machine Learning | PyTorch, Scikit-learn, RedisAI, RCNN, Computer vision pipelines, Video fragmentation and scene analysis |
Backend | Python , Django |
Frontend | React.js |
Databases | PostgreSQL, MongoDB, InfluxDB, RDS |
Cloud and Infrastructure | AWS Amplify, AWS Lambda, AWS EC2 (G-type instances) |
Impact: Smarter Content, More Effective Advertising
The platform delivered measurable improvements
Segment-level understanding of video content
Improved advertiser performance through contextual targeting
A system that improves as usage grows
Higher relevance in ad delivery without relying solely on user data
Support for both mass-market and niche campaigns
Why This Matters
OTT advertising does not fail due to lack of data.
It fails due to lack of context.
By teaching the system to understand visuals, sound, mood, and behavior together, the platform moved beyond generic recommendations toward intelligent decision-making.
This created a foundation for long-term value, not just short-term optimization.
Beyond Launch: Built to Scale and Learn
The intelligence system continues to evolve through
Improved semantic models
Expanding information architecture
More accurate scene interpretation
Stronger advertiser feedback loops