What goes in, what comes out
IN
Raw or partial data at any stage of the pipeline: incoming schedules, unmatched records, untagged images, untranslated descriptions.
OUT
AI-assisted output passed to editorial review: matched records with confidence scores, drafted enrichment, flagged anomalies, ready for an editor to confirm, correct or override.
What the AI does
Visual AI works on every image so it arrives ready to display at any crop and resolution:
- tags content and detects faces and objects (around a dozen neural networks run on each photo)
- picks the main subject to drive smart cropping
- corrects colour automatically
- upscales lower-resolution assets
- scores each image for quality and relevance
Text AI turns source data into clean, consistent text:
- drafts programme synopses from structured data and source text
- translates metadata across many languages
- condenses descriptions to the exact length a channel needs
- extracts genre, thematic and intent tags
- applies content-safety tagging (age-appropriateness, violence, nudity)
- removes duplicate descriptions
Matching AI links each incoming broadcast to the correct master record, so the same title stays consistent across every channel and source. This is the work that makes one permanent ID possible at scale.
Search and discovery AI generates semantic embeddings for titles, so content can be found by meaning rather than exact keywords, and surfaces correctly in modern, AI-driven search. These are regenerated automatically whenever a record changes. We supply this layer as structured data and embeddings; your own discovery and recommendation logic stays yours.
Editors stay in control
Every AI output is reviewed by a human editor before it reaches you. No field is AI-only: the AI proposes, an editor decides. Each output also carries:
- full provenance: which model produced it, when, and with what confidence
- a separate log of the editorial review, kept distinct from the AI source
- the ability to roll back changes
The AI is also deliberately constrained. For text, the models are restricted to extracting and summarising from verified source data, not inventing from their own training, which is what keeps enrichment grounded and guards against hallucination. On top of that sit automated schema and confidence checks, drift monitoring, the mandatory human review, and ongoing regression testing. The principle is the same across the platform: automation where it is safe, human oversight where it matters.
What matters is the data, not who made it
AI here is a means, not a selling point in itself. A description is not better because a model wrote it, and not worse because an editor did; what matters is that the record is accurate, complete and consistent. The AI Suite earns its place by doing the repetitive work at scale so editors can spend their judgement where it counts, and by making sure every output is checked before it reaches you.
Built and run in the EU
The AI models run on Media Press's own GPU infrastructure inside the EU. For European broadcasters and operators, that means enrichment at scale without sending content to a third-party cloud:
- customer data does not leave the EU (on-premise models give full data sovereignty)
- AI use is classified as low-risk under the EU AI Act (metadata enrichment only)
- full model documentation, mandatory human review gates and immutable audit logs
- where high-complexity tasks use external frontier models, they run under GDPR-compliant data-processing agreements
Where it works
The AI Suite cuts across all four modules of the MP Platform: visual AI in the Media & Image Hub, text and matching AI in Metadata Production, and semantic search across the platform, supporting every product it delivers.
Products that live here
Get started
See the AI Suite in action
Book a demo to see the AI Suite in action on your data.