# Core Architecture

**1. Data Layer**

<figure><img src="https://292182233-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FYEAdPOy53SCSvHnNXMEm%2Fuploads%2FxUhVhuky4MhwqRvg6Nd2%2FSoundr%20Core%20Architecture%20(1).png?alt=media&#x26;token=bf002ef9-5ada-4750-8340-0d3d7a80e956" alt=""><figcaption><p>Soundr Core Architecture</p></figcaption></figure>

We collect and index from diverse, high-volume and high-entropy content sources, including:

* Twitter/X firehose (with semantic deduplication)
* Forum scrapers (Discourse, Commonwealth, Snapshot, etc.)
* Academic papers & technical reports
* GitHub & open-source code updates
* Podcasts (transcribed)
* Event transcripts (ETHCC, Devcon, etc.)
* Media outlets, substack, and newsletters

All data is converted into a standardized format (JSON-LD), tokenized, and stored in a time-versioned vector database.

**2. AI Engine**

Our intelligence stack includes:

* **Transformer-based NLP models** (custom fine-tuned LLMs for crypto context)
* **Multi-modal embeddings** for text + audio (Whisper + BERT hybrid stack)
* **Graph analytics** to connect authors, narratives, and influence nodes
* **Topic modeling** to extract dominant themes and trending subcultures
* **Mindshare computation**: measures content impact using weighted interaction, temporal resonance, and project alignment

We continuously fine-tune our AI on millions of crypto-native datapoints, ensuring context specificity

#### 3. **Application Layer**

This is where the user meets the intelligence. Key interfaces include:

* **Soundr Dashboard** – Unified search & trend analysis panel
* **Soundr Leaderboards** – Real-time, campaign-driven ranking system
* **Soundr Points (SUPs)** – Reward layer built atop community interaction signals
* **Public API** – For project teams, analysts, and researchers to access structured, filtered data