flowchart LR
    subgraph Reactive["🛡️ Reactive Detection"]
        R1["User sends content\n(text / image / audio / video)"]
        R2["AI-generation verdict\n+ confidence score\n+ spoken explanation"]
        R1 --> R2
    end

    subgraph Proactive["🔮 Proactive Prediction"]
        P1["Comms officer pastes\nannouncement draft"]
        P2["Predicted false narratives\n+ counter-narratives in 4 languages\n+ one-click deployment"]
        P1 --> P2
    end

    style Reactive fill:#1e1b4b,stroke:#6366f1,color:#c7d2fe
    style Proactive fill:#14532d,stroke:#22c55e,color:#bbf7d0

flowchart TB
    subgraph Entry["Entry Points"]
        TG["Telegram Bot\n/detect · /predict · media"]
        WEB["Web Dashboard\nSSE streaming · WebSocket audio"]
    end

    TG & WEB --> Router

    subgraph Router["Request Router"]
        direction LR
        DET["Detect\n(text/image/audio/video)"]
        PRED["Predict\n(announcement)"]
    end

    DET --> Reactive
    PRED --> Proactive

    subgraph Reactive["Reactive Pipeline"]
        G["GUARD\nSEA-LION"]
        M["Misinfo\nGemini"]
        IM["Manipulation\nOpenCV"]
        INS["Insights\nGemini → Groq"]
        G & M & IM --> INS
    end

    subgraph Proactive["Proactive Pipeline"]
        TE["Topic Extraction\nGemini structured JSON"]
        SR["Source Retrieval\n(parallel)"]
        FC["Firecrawl\nPOFMA · CNA · MOH"]
        RAG["ClickHouse RAG\nHistorical vectors"]
        RP["Rumour Prediction\nGemini structured JSON"]
        TE --> SR
        SR --> FC & RAG
        FC & RAG --> RP
    end

    Reactive --> Reply
    Proactive --> Reply

    subgraph Reply["Response"]
        TR["Translation\nSEA-LION Gemma"]
        FMT["Formatter\nHTML"]
        LIVE["Gemini Live API\nSpoken verdict"]
        DEPLOY["Telegram Deploy\nCounter-narratives"]
        TR --> FMT & LIVE & DEPLOY
    end

    Reply --> LOG["ClickHouse\nTelemetry + RAG"]

    style Entry fill:#1e1b4b,stroke:#6366f1,color:#c7d2fe
    style Reactive fill:#172554,stroke:#3b82f6,color:#bfdbfe
    style Proactive fill:#14532d,stroke:#22c55e,color:#bbf7d0
    style Reply fill:#3b0764,stroke:#a855f7,color:#e9d5ff

flowchart LR
    A["📱 User sends\ncontent"] --> B["🌐 Language\nDetection"]
    B --> C["🔄 Translate\nto English"]
    C --> D["⚡ Parallel Detection"]

    subgraph D["Parallel Detection"]
        D1["GUARD\n(AI safety)"]
        D2["Misinfo\n(LLM analysis)"]
        D3["Manipulation\n(OpenCV)"]
    end

    D --> E["💡 Insights\n(LLM explanation)"]
    E --> F["🔄 Translate\nback"]
    F --> G["📨 HTML verdict\n+ voice note"]

    style D fill:#172554,stroke:#3b82f6,color:#bfdbfe

flowchart TD
    A["📋 Announcement draft"] --> B["🔍 Extract Topics\ntopics · communities · triggers"]
    B --> C{"Retrieve Sources\n(parallel)"}
    C --> D["🌐 Firecrawl\nPOFMA · CNA · MOH"]
    C --> E["📚 ClickHouse RAG\nHistorical articles"]
    D & E --> F["🧠 Predict Rumours\n3-8 predictions ranked by risk"]
    F --> G["📊 Results"]

    subgraph G["Rumour Forecast"]
        G1["🔴 CRITICAL — Rice shortage rumour\n📱 WhatsApp Mandarin · ~2hrs"]
        G2["🟠 HIGH — Toilet paper panic\n📱 Cross-language WhatsApp · ~4hrs"]
        G3["🟡 MEDIUM — Government hiding cases\n📱 Twitter/Reddit · ~6hrs"]
    end

    G --> H["✅ Deploy counter-narratives\nEN · 中文 · BM · தமிழ்\n→ 800+ community leaders"]

    style G1 fill:#7f1d1d,stroke:#dc2626,color:#fca5a5
    style G2 fill:#7c2d12,stroke:#ea580c,color:#fed7aa
    style G3 fill:#713f12,stroke:#ca8a04,color:#fef08a
    style H fill:#14532d,stroke:#22c55e,color:#bbf7d0

flowchart LR
    A["📄 Announcement"] --> B["Gemini Embedding\n768-dim vector"]
    B --> C["Phase 1\nTopic Filter\n+ cosineDistance"]
    B --> D["Phase 2\nPure Vector\ncosineDistance"]
    C & D --> E["Dedup & Merge\nTopic-matched first"]
    E --> F["Credibility-weighted\nRAG Sources"]

    style A fill:#1e1b4b,stroke:#6366f1,color:#c7d2fe
    style F fill:#14532d,stroke:#22c55e,color:#bbf7d0