07. Build an AI Document Assistant

flowchart TD
    subgraph FRONT["Frontend"]
        UI["Document Library"]
        VIEWER["Document Viewer\nPDF + Annotations"]
        SEARCH["Search + Q&A"]
    end
    subgraph PROCESS["Processing Pipeline"]
        INGEST["Ingestion Service"]
        OCR_SVC["OCR Service\nTesseract + Document AI"]
        CHUNK["Chunking\nSemantic splitting"]
        EMBED["Embeddings\ntext-embedding-3-small"]
        INDEX["Vector Index\nQdrant"]
    end
    subgraph SERVICES["AI Services"]
        QA["Q&A Service\nRAG pipeline"]
        COMPARE["Document Comparison"]
        SUMMARIZE["Summarization"]
        TRANSLATE["Translation"]
    end
    subgraph DATA["Storage"]
        PG["PostgreSQL\nMetadata"]
        VDB["Qdrant\nVectors"]
        S3["Document Store\nPDF/Images"]
        REDIS["Redis\nCache"]
    end

    UI --> INGEST
    INGEST --> OCR_SVC
    OCR_SVC --> CHUNK
    CHUNK --> EMBED
    EMBED --> INDEX
    INGEST --> S3
    CHUNK --> PG
    SEARCH --> QA
    QA --> VDB
    QA --> S3

    style FRONT fill:#3b82f6,color:#fff
    style PROCESS fill:#f59e0b,color:#fff
    style SERVICES fill:#22c55e,color:#fff
    style DATA fill:#6366f1,color:#fff

sequenceDiagram
    participant U as User
    participant API as API
    participant Queue as Queue
    participant OCR as OCR Worker
    participant Chunk as Chunker
    participant VDB as Vector DB

    U->>API: Upload scanned PDF
    API->>Queue: Enqueue processing job
    API-->>U: Processing started (job_id)

    Queue->>OCR: Process document
    OCR->>OCR: OCR scan pages
    OCR-->>Chunk: Extracted text

    Chunk->>Chunk: Split into semantic chunks
    Chunk->>Chunk: Generate embeddings per chunk
    Chunk->>VDB: Store vectors + metadata
    Chunk-->>API: Processing complete

    API-->>U: Document ready (WebSocket update)

    U->>API: "What are the key terms in this contract?"
    API->>VDB: Search similar chunks
    VDB-->>API: Top-5 relevant chunks
    API->>LLM: Generate answer with citations
    LLM-->>API: Answer with page references
    API-->>U: Display answer with source highlights

flowchart TD
    QUERY["User Question"] --> EMBED_Q["Embed Question"]
    EMBED_Q --> SEARCH_V["Search Vector DB\nTop-K chunks"]
    SEARCH_V --> RE_RANK["Re-rank\nCross-encoder scoring"]
    RE_RANK --> TOP_K["Top 3-5 chunks"]
    TOP_K --> BUILD_CONTEXT["Build Prompt\nQuestion + Context"]
    BUILD_CONTEXT --> LLM["LLM Generate Answer\nWith source citations"]
    LLM --> CITE["Verify Citations\nAre sources accurate?"]
    CITE --> RESPONSE["Return Answer\n+ Document links"]

    style QUERY fill:#f59e0b,color:#fff
    style SEARCH_V fill:#3b82f6,color:#fff
    style LLM fill:#22c55e,color:#fff

flowchart LR
    DOC_A["Document A\nVersion 1"] --> PARSE_A["Parse\nExtract clauses"]
    DOC_B["Document B\nVersion 2"] --> PARSE_B["Parse\nExtract clauses"]
    PARSE_A --> ALIGN["Align Sections\nSame clause matching"]
    PARSE_B --> ALIGN
    ALIGN --> DIFF["Identify Changes\nAdded, removed, modified"]
    DIFF --> CATEGORIZE["Categorize Changes\nMinor, significant, critical"]
    CATEGORIZE --> REPORT["Comparison Report\nSide-by-side diff"]

    style DIFF fill:#3b82f6,color:#fff
    style REPORT fill:#22c55e,color:#fff