23 Maestro engines.Zero manual tagging.

One brain for every LLM call.

Unified Vertex Gemini routing with optional Bedrock fallback, containment prefix on all system prompts, and credit metering after successful completions.

• Single LLM entry point
• Configure-mode when Vertex is off
• Bill after isBillableLlmResult()

Layer 2

Kill switch for auto-executing AI.

Platform-wide guard that blocks autonomous actions like Cortex goal decomposition until operators explicitly enable them.

• Ops-controlled autonomy
• Gates Cortex decompose
• Zero credit cost

Layer 2

Every decision recorded for tuning.

Captures Maestro decision outcomes so prompts and models can improve over time without losing auditability.

• PostgreSQL decision log
• Feeds staff ops review
• Future prompt tuning

Layer 2

BPM and beat grid, solved locally.

RhythmExtractor2013 for tempo plus madmom DBN beat tracking for complex rhythms. Powers auto-trim and catalog search filters.

• >95% BPM accuracy target
• Beat timestamps for editors
• $0 local inference

Layer 1

Tags every upload automatically.

Auto-tags mood, energy, and genre from raw audio so partners can search by feel without manual metadata entry.

• Feeds recommendations
• Brand match signals
• No LLM on hot path

Layer 1

Duplicate and rights signals.

Chromaprint (fpcalc) fingerprints flag tracks highly similar to existing catalog entries before they reach partners.

• AcoustID-compatible
• Duplicate detection
• Local binary

Layer 1

Lyrics screened before publish.

Two-stage pipeline: transcribe vocals, then classify toxicity. Flags content for human review instead of silent auto-approve.

• Explicit content screening
• Staff review queue
• Local or API Whisper

Layer 1

Bios, descriptions, press releases.

Generates artist bios, track descriptions, and press copy with catalog context. Advisory output: artists review before publishing.

• Credit-metered actions
• Context enrichment first
• 10-80 credits per action

Layer 2

Your AI studio partner.

Multi-session chat for pitch emails, sync angles, and captions. Persisted threads in the artist portal at /ai-studio.

• Chat, Bio, Fan 360 tabs
• 2 credits per message
• Catalog-aware prompts

Layer 2

Multi-language catalog copy.

Translates bios and metadata for international partner surfaces without duplicate manual entry.

• 8 credits per run
• Locale-aware
• Artist review recommended

Layer 2

Clips at musical phrases.

Finds preview clip boundaries at downbeats and low-energy transitions using beat grids from the sidecar.

• No LLM latency
• Phrase-aligned previews
• Human feedback path

Layer 2

Artists bring own artwork.

Cover art generation is intentionally deferred. Maestro focuses on audio intelligence and copy; artists upload artwork.

• Not in scope today
• Manual cover upload
• See artist portal

Layer 2 · deferred

Five-signal scoring, config-driven.

Content similarity, collaborative filtering, contextual boost, popularity decay, and diversity penalty. Weights from recommendation_surface_config, never hardcoded.

• For You shelf
• Co-usage matview
• ~$0.001 per query

Layer 2

LLM tags on sidecar metadata.

Layers LLM-suggested genre and mood tags on top of deterministic audio classifiers for richer partner search.

• 3 credits per enrich
• Sidecar + LLM hybrid
• Track catalog feed

Layer 2

Every LLM call gets context.

Aggregates platform usage, catalog health, seasonal signals, and artist insights before Maestro inference.

• getContext() before LLM
• Shared across plugins
• Optional Haiku expansion

Layer 2

Commerce-ready mood fit.

Maps brand categories to mood and energy profiles, evolving toward cross-modal embedding similarity for ad placements.

• Phase 1 lookup table
• Phase 2 learned match
• Partner campaign fit

Layer 2

Every listener gets For You.

Taste vectors from likes, skips, playlist adds, and unlikes. Refreshed periodically; powers the Explore For You shelf.

• behavioral_event loop
• GDPR-aware listener scope
• No LLM on hot path

Layer 2

Know who engages with your catalog.

Segments engaged listeners and ranks tracks by affinity for artist dashboard insights in AI Studio.

• fan_track_affinity MV
• Artist portal tab
• Behavior-driven refresh

Layer 2

Listener help on autopilot.

Answers discovery, library, and licensing basics at /support. Escalates to human support when needed.

• 1 credit per message
• 50 starter listener credits
• Containment rails

Layer 2

Business goals → task lists.

Artists define goals; Maestro decomposes them into actionable tasks gated by the autonomy engine.

• 20 credits per decompose
• API live in portal
• BCC-adjacent, not BCC

Layer 2

Catalog context for platforms.

Partner portal POST enrich turns track summaries into structured enrichment for internal workflows.

• 15 credits per call
• 200 platform starter credits
• OpenAPI-documented

Layer 2

Heuristic sync price ranges.

Suggests sync pricing from license history with optional LLM narrative for artist review.

• 5 credits when LLM used
• Advisory only
• Artist portal API

Layer 2

Release date candidates.

Friday release suggestions plus seasonal context from catalog and platform patterns.

• 5 credits when LLM used
• Seasonal SQL signals
• Artist dashboard API

Layer 2

Internal ops assistant.

Admin panel chat with tool context for pipeline health, moderation backlog, and platform metrics.

• 1 credit after success
• 10k staff starter credits
• Rate limited

Layer 2