Pod is a media and web service where music is curated not by genre or artist, but by life situation. The user describes their moment using simple parameters—task, energy level, presence of vocals, time of day, weather, and emotion—and receives a ready-made playlist that can be opened in a streaming service with a single click.

At the core of the platform is a generative pipeline: AI agents gather sources and draft scenarios, an editor controls quality and publication, and the community validates the outcome through a «yes/no» vote. This allows Pod to eliminate the chaos of endless genre suggestions and provides confidence that the music truly fits the specific moment.

The primary goal of Подо is to make music discovery practical and fast, reduce the time spent browsing tracks, and create a space where people can share their musical choices and inspire one another.

For «Pod, ” which curates music based on scenarios, it is critically important to separate the sources into those from which the AI gathers „raw material“ for analysis, and those from which it takes the tracks themselves.

Since «Podo» does not host music itself but functions as an aggregator, we require access to the catalogs and metadata.Streaming Platform APIs (Yandex Music, Spotify): These are used to build the on-site player and verify that a track is available for playback.Open Metadata Databases (MusicBrainz, Discogs): AI analyzes this data to determine if a track matches the user’s request.Audio Analysis (Proprietary algorithms or services): We use this to provide the user with an accurate «Energy 4/10» slider. The AI must analyze the audio file itself, not just the tags.

This is one of the most valuable points because it gives «Pod» uniqueness—social validation.Pod’s internal database (proprietary UGC): AI analyzes which words are most frequently used under specific playlists, and how people describe their emotions. Voting shapes the credibility rating of the scenario.Comments and discussions (Telegram, VK, music forums): NLP models extract «real-life» scenarios and associations—"situation = music"—from these sources.User playlists (from streaming services and social networks): This is a ready-made dataset for training AI to understand what music people choose for everyday tasks.

For automatic parameter substitution to function, we need data about the world.

Weather APIs (Yandex Weather): If it is raining outside, the site can automatically suggest the scenario «Rain Outside.»

Time and Calendar: On Monday mornings, we suggest «Get Ready for Work, ” and on Friday evenings— „Going Home.“

Geolocation and Map APIs (Yandex Maps): We use these to provide scenarios such as „On the Road“ or localized B2B scenarios.

A person in the pipeline must have something to anchor them while reviewing AI drafts.Music press and blogs (Pitchfork, The Blueprint): The editor reads how critics describe the track to determine if it truly fits the stated scenario.Music influence research (scientific databases, Yandex Studies): We use this to create expert content («Moment Breakdowns») and to fine-tune algorithms.

To understand which scenarios will be popular tomorrow.Search analytics services (Yandex.Wordstat, Google Trends): We create scenarios based on real demand. If we see an explosive growth in the search query «music for meditation,» we launch that scenario.Trend watching (social media): We use this to create relevant visual covers and themed collections.

1. Data Collection Machine (AI): Scans streaming APIs (Yandex, Spotify), metadata databases (MusicBrainz), social networks, and weather services.Human: Approves the «whitelist» of reliable sources and defines stop words (e.g., banning pirated websites).

2. Analysis and Filtering Machine (AI): Analyzes audio (detects BPM, energy, vocals), filters out duplicates, and automatically groups tracks based on parameters.Human: Reviews ambiguous tracks that the algorithm could not classify definitively and fine-tunes parameter priorities.

3. Generation Machine (AI): Assembles the draft script: writes the text, forms a tracklist of 20–30 songs, and selects a cover art.Human: Sets the general theme for the week, approves the Tone of Voice, and establishes editorial guidelines.

4. Verification and Publication Machine (AI): Checks that all track links are functional and available across streaming platforms.Human: Selectively listens to the playlist, verifies the song lyrics, and presses the «Publish» button.

5. Feedback Machine (AI): Gathers statistics—how many people clicked «attend / not attend,» how many listened.

Human: Reads user comments, draws conclusions about why the scenario might have failed, and adjusts the rules for the AI.

Task: Scan the internet and streaming services to find new scenarios and tracks.

Whitelist Rule: The Agent is authorized to access only those APIs and websites approved by the user (Yandex Music, MusicBrainz, official music media outlets).

Piracy Prohibition: Any attempt by the AI to scrape data from unlicensed websites or torrents is blocked.

Anonymization: When collecting comments and user playlists from social networks, the AI must remove personal data (names, geolocation), retaining only the text.

Limits: The Agent cannot make more than N requests per minute to avoid overloading partner servers.