VibeGraph AI
— The Audiophile's Acoustic Signal Linker
Live Web App: vibe-graph-ai.vercel.app
What it does:
VibeGraph AI is a graph-native recommendation engine built for audio enthusiasts. Instead of relying on mass-market collaborative filtering (suggesting what other people clicked), it uses pure digital signal processing and structural graph musicology to recommend tracks based on their actual physical sound waves.
If a user wants to test a new pair of IEMs or active bookshelf speakers with a specific gritty, raw track, VibeGraph AI finds other music that hits those same frequencies—even across entirely different genres. It does this by analyzing a 33-Dimensional Acoustic Fingerprint for every track:
-
Timbral Envelope (20 MFCCs): The raw texture of the instruments.
-
Harmonic Footprint (12 Chroma): The underlying pitch and key signatures.
-
Spectral Brightness (1 Centroid): The perceived treble energy.
The Zero-Hallucination Design:
The core of VibeGraph AI is its deterministic mathematical restraint.
-
If asked to find an Electronic match for a heavy Punk Rock track, the agent analyzes the pre-calculated acoustic neighbors.
-
If the physical math does not intersect with the requested genre, it refuses to hallucinate a fake recommendation. It respects the strict boundaries of the graph, explaining the anomaly to the user instead of lying to fulfill the prompt.
The Tools Powering the Agent:
-
Pre-Computed Acoustic Topology: Instead of relying purely on expensive runtime vector searches, we pre-calculated the top 5 nearest mathematical neighbors for all 25,000 tracks. The agent natively traverses explicit
(Track)-[:SOUNDS_LIKE {score}]->(Track)edges for lightning-fast recommendations. -
Semantic Vector Search: For completely novel queries, the agent can fall back to performing a native cosine similarity search across the 33-dimensional embedding matrix.
-
Metadata Graph Explorer (Text2Cypher): Dynamically filters acoustic matches by traversing the graph topology along
(Track)-[:IN_GENRE]->(Genre)pathways. -
Acoustic Landscape Analyzer: A Cypher Template that calculates structural clusters and track catalog densities on demand.
Dataset and why a graph fits:
-
The Dataset: Free Music Archive (FMA) Medium — 25,000 tracks.
-
Why a graph is required: Acoustic similarity matching on its own is an isolated vector space challenge. However, music discovery is a multi-signal network challenge. A track's mathematical audio coordinates must intersect with historical eras, artist identifiers, and genre taxonomies.
By hardcoding the 33D DSP nearest-neighbor vectors into [:SOUNDS_LIKE] edges directly between Track nodes, the Aura Agent can run multi-hop queries that traverse boundaries to find hidden relational structures. A flat vector database cannot trace these relationships, and a standard SQL table cannot perform the vector math. The graph allows the agent to reason over both simultaneously.
Links & Code:
- GitHub Repository (Agent Config & Pipeline): https://github.com/Shivam007kumar/VibeGraph-AI
Screenshots of VibeGraph AI in Action:
