Spotify / Signal.
The next discovery, with more context.
An idea.
Another perspective.
Independent discovery concept for creators inspired by Spotify. Semantic search, explainable recommendations and tools to connect audiences with new sounds.
The challenge
We imagine a catalog with incomplete metadata, a long tail of hard-to-discover creators, and recommendations that need to respect preferences and privacy. The challenge is improving discovery without locking the audience into the familiar.
The proposed direction
We propose a hybrid text and vector index, metadata enrichment with human review, and a ranking that balances relevance, novelty and diversity. A cohort lab would separate correlation from impact and allow disabling models that worsen the experience.
Architecture and technologies
Engineering challenges and decisions
- Popularity bias: explicit diversity and cohort evaluation.
- Privacy: data minimization and consent controls.
- Inaccurate responses: catalog references and editorial evaluation.
Modeled advertising revenue
Concept projection, calculated from the following assumptions.
Visible assumptions
20 million ad opportunities/month and 20 USD fixed net CPM. Baseline: 52% fill rate. Hypothesis: 66% fill rate. Exploration model assumptions.
20.000.000 ÷ 1.000 × 20 × 0,52 = 208.000 USD → × 0,66 = 264.000 USDHow we would validate it
Validate fill rate, retention and ad load with an experiment. Do not attribute the increase to AI without causal evidence or confuse revenue with profit.
A conversation can open up a whole universe.