flowchart TD
A(["🌐 RSS Feeds — BBC · NYT · Al Jazeera"]) -->|feedparser| B(["📰 Raw Articles"])
B -->|TF-IDF vectorization| C(["🔢 Feature Vectors"])
C -->|DBSCAN clustering| D{{"🗂️ Cluster? — similarity > 0.3"}}
D -- "Multi-source" --> E(["🧠 LLM Extraction — Llama 3.2"])
D -. "Single article" .-> F(["📝 Standalone Article"])
E --> G(["✅ Verified Facts + Classification"])
G -->|geography · category| H[("🗄️ SQLite DB")]
F -.-> H
H -->|parent / child links| I(["🔗 DAG Builder"])
I --> J(["📊 Knowledge Graph — vis-network"])
H --> K(["🖥️ Dashboard — Filterable UI"])
style A fill:#312e81,stroke:#818cf8,stroke-width:2px,color:#e0e7ff
style B fill:#1e1b4b,stroke:#6366f1,stroke-width:2px,color:#c7d2fe
style C fill:#1e1b4b,stroke:#6366f1,stroke-width:2px,color:#c7d2fe
style D fill:#3b0764,stroke:#c084fc,stroke-width:2px,color:#f3e8ff
style E fill:#312e81,stroke:#818cf8,stroke-width:2px,color:#e0e7ff
style F fill:#1e293b,stroke:#475569,stroke-width:1px,color:#94a3b8
style G fill:#064e3b,stroke:#34d399,stroke-width:2px,color:#d1fae5
style H fill:#1e1b4b,stroke:#a78bfa,stroke-width:2px,color:#ddd6fe
style I fill:#312e81,stroke:#818cf8,stroke-width:2px,color:#e0e7ff
style J fill:#1e1b4b,stroke:#6366f1,stroke-width:2px,color:#c7d2fe
style K fill:#1e1b4b,stroke:#6366f1,stroke-width:2px,color:#c7d2fe
linkStyle default stroke:#818cf8,stroke-width:2px
linkStyle 5 stroke:#475569,stroke-width:1px,stroke-dasharray:6
linkStyle 8 stroke:#475569,stroke-width:1px,stroke-dasharray:6
1. RSS Ingestion
Fetches live feeds from global sources using feedparser. Collects titles, descriptions,
and publication dates.
2. Semantic Clustering
Vectorizes text with TF-IDF and groups similar articles using DBSCAN
(cosine similarity > 0.3).
3. LLM Extraction
Passes clusters to Llama-3.2 to synthesize titles, extract verified facts, and classify
by geography & category.
4. DAG Generation
Links events chronologically (parent → child) and stores them in SQLite. The Knowledge Graph renders
this as a directed, hierarchical timeline.