WYnova

SAPO: AI content recommendation and audio

WYnova developed an AI recommendation engine, text-to-speech audio and automatic categorisation with 90% accuracy for SAPO, under human validation by the newsroom.

Client
SAPO
Sector
Media
Technology
LLM, PGVector, text-to-speech
Recommendation
Pilot Q2 2025
Contact WYnova
SAPO's AI content organisation and recommendation platform
Mock-up of the SAPO content platform: recommendations and audio
90%

accuracy in categorisation

Pilot Q2 2025

recommendation engine

Audio

via integrated text-to-speech

The challenge

SAPO's editorial processes limited personalisation and accessibility. Content discovery was static, with the same feed for every reader, and the volume of articles made manual curation unsustainable.

Key obstacles

  • Manual categorisation

    Inconsistent classification, unsustainable at that volume.

  • No customisation

    The same static feed for all readers.

  • Limited formats

    Content limited to text, with no audio or summaries.

The solution

WYnova developed a recommendation engine, text-to-speech audio generation and automatic article categorisation. All components operate under newsroom validation.

SAPO · recomendaçãoPilot

Which articles should I recommend to anyone who has just read this one?

Suggestion from the recommendation engine

Three articles on related topics, sorted by similarity. The tags for each article have been approved by the editorial team.

Calculated similarity using PGVector embeddings

  1. Related article A · 2 tags in common
  2. Related article B · 2 tags in common
  3. Related article C · 1 tag in common
Illustration of the recommendation engine developed for SAPO. Articles are illustrative.
  1. 01

    Contents

    Ingestion of the editorial catalogue and its metadata.

  2. 02

    AI categorisationAI

    Automatic categorisation with an LLM, validated by the newsroom.

  3. 03

    Embeddings

    Semantic indexing with PGVector for search and links between articles.

  4. 04

    RecommendationAI

    Recommendation engine using LLMs and embeddings, personalised for each reader.

  5. 05

    Audio (TTS)AI

    Text-to-speech and automatic summaries for accessibility.

  6. 06

    Curation

    Human validation: the newsroom approves and retains editorial control.

AI broadens content discovery at SAPO, while editorial decisions remain with the newsroom.

Results

90% categorisation accuracy, a recommendation engine piloted in Q2 2025 and integrated audio. The editorial team has redeployed time to higher-value work.

What we learned

In a media organisation, AI adds scale to discovery and accessibility when editorial judgement remains with the newsroom.

Related cases and services

More cases and solutions in AI for the media and audiovisual sector.

Auchan

AI-automated promotional leaflets: +25% monthly savings.

QNB

Requirements process redesigned across IT, Business and Product.
Service

AI Solution Builder

Case studies

View all cases

Content discovery at the scale of your catalogue

WYnova identifies where AI can accelerate the discovery and accessibility of your organisation's content.