Merge pull request #1 from noelty/dev

app.py

@@ -0,0 +1,77 @@
+import os
+import nltk
+import gradio as gr
+from documents import process_docx, process_pdf, process_txt
+from indexing import index_document
+from querying import query_documents
+import preprocess
+
+# Download required NLTK data (do this *once* when the app starts)
+try:
+    nltk.data.find("corpora/wordnet")
+except LookupError:
+    nltk.download("wordnet")
+
+try:
+    nltk.data.find("corpora/stopwords")
+except LookupError:
+    nltk.download("stopwords")
+
+try:
+    nltk.data.find("tokenizers/punkt")
+except LookupError:
+    nltk.download("punkt")
+
+UPLOAD_FOLDER = 'uploads'
+os.makedirs(UPLOAD_FOLDER, exist_ok=True)
+
+
+def process_and_query(file, query_text):
+    """
+    Processes a document, indexes it, and performs a query.  This is the
+    main function called by the Gradio interface.
+    """
+    if not file:
+        return "No file uploaded", []
+
+    file_path = file.name  # Gradio passes a NamedTemporaryFile
+
+    # Process file
+    if file.name.endswith('.docx'):
+        text = process_docx(file_path)
+    elif file.name.endswith('.pdf'):
+        text = process_pdf(file_path)
+    elif file.name.endswith('.txt'):
+        text = process_txt(file_path)
+    else:
+        return "Unsupported file type", []
+    preprocessed_text = preprocess.preprocess_text(text['text'])
+    print (preprocessed_text) #ADD THIS
+
+    # Index the document
+    index_result = index_document("documents", file.name, preprocessed_text)
+
+    # Perform the query
+    query_results = query_documents("documents", query_text)
+
+    return f"Indexing result: {index_result}", query_results
+
+
+# Gradio Interface
+iface = gr.Interface(
+    fn=process_and_query,
+    inputs=[
+        gr.File(label="Upload Document"),
+        gr.Textbox(label="Enter Query")
+    ],
+    outputs=[
+        gr.Textbox(label="Indexing Result"),
+        gr.JSON(label="Query Results") # Display query results as JSON
+    ],
+    title="Document Processing and Query",
+    description="Upload a document (docx, pdf, or txt), enter a query, and get the results."
+)
+
+
+if __name__ == '__main__':
+    iface.launch()

collection.py

@@ -0,0 +1,32 @@
+from qdrant_client import QdrantClient
+from qdrant_client.http.models import VectorParams, Distance
+import os
+
+# Connect to the local Qdrant instance (using environment variables)
+QDRANT_HOST = os.environ.get("QDRANT_HOST", "localhost")
+QDRANT_PORT = int(os.environ.get("QDRANT_PORT", 6333))
+
+client = QdrantClient(host=QDRANT_HOST, port=QDRANT_PORT)
+
+# Define collection name and vector parameters (using environment variables)
+COLLECTION_NAME = os.environ.get("QDRANT_COLLECTION_NAME", "documents")
+VECTOR_SIZE = int(os.environ.get("QDRANT_VECTOR_SIZE", 384))  # Adjust based on your embeddings
+
+#Map the string to the Distance Enum.
+DISTANCE_METRIC_STRING = os.environ.get("QDRANT_DISTANCE_METRIC", "Cosine").lower()
+DISTANCE_METRIC = Distance.COSINE
+if(DISTANCE_METRIC_STRING == "euclid"):
+    DISTANCE_METRIC = Distance.EUCLID
+elif(DISTANCE_METRIC_STRING == "dot"):
+    DISTANCE_METRIC = Distance.DOT
+
+# Create the collection
+try:
+    client.recreate_collection(
+        collection_name=COLLECTION_NAME,
+        vectors_config=VectorParams(size=VECTOR_SIZE, distance=DISTANCE_METRIC),
+    )
+    print(f"Collection '{COLLECTION_NAME}' created/recreated successfully!")
+
+except Exception as e:
+    print(f"Error creating/recreating collection '{COLLECTION_NAME}': {e}")

documents.py

@@ -0,0 +1,39 @@
+import docx
+import fitz  # PyMuPDF
+
+def process_docx(file_path):
+    """Extracts text from a .docx file."""
+    try:
+        doc = docx.Document(file_path)
+        full_text = [para.text for para in doc.paragraphs]
+        text = '\n'.join(full_text)
+        
+        print(f"Extracted {len(full_text)} paragraphs from DOCX")  # Debugging
+        print(f"Extracted Text: {text[:500]}...")  # Print first 500 chars
+        
+        return {'text': text.strip()}
+    except Exception as e:
+        return {'error': str(e)}
+
+
+def process_pdf(file_path):
+    """Extracts text from a .pdf file."""
+    try:
+        pdf = fitz.open(file_path)
+        text = ""
+        for page in pdf:
+            text += page.get_text()
+        pdf.close()
+        return {'text': text.strip()}  # Return as a dictionary
+    except Exception as e:
+        return {'error': str(e)}
+
+
+def process_txt(file_path):
+    """Extracts text from a .txt file."""
+    try:
+        with open(file_path, 'r', encoding='utf-8') as f:
+            text = f.read()
+        return {'text': text.strip()}  # Return as a dictionary
+    except Exception as e:
+        return {'error': str(e)}

indexing.py

@@ -0,0 +1,128 @@
+import uuid
+import re
+import logging
+import nltk
+from qdrant_client import QdrantClient
+from qdrant_client.http.models import VectorParams, Distance
+from sentence_transformers import SentenceTransformer
+
+# Download tokenizer for sentence splitting
+nltk.download("punkt")
+from nltk.tokenize import sent_tokenize
+
+# Initialize Qdrant client and model
+qdrant_client = QdrantClient(host="localhost", port=6333)
+model = SentenceTransformer('all-MiniLM-L6-v2')
+
+# Set up logging
+logging.basicConfig(level=logging.INFO)
+
+def create_collection_if_not_exists(collection_name):
+    """Creates a Qdrant collection if it doesn't already exist."""
+    try:
+        collections_response = qdrant_client.get_collections()
+        existing_collections = [col.name for col in collections_response.collections]
+
+        if collection_name not in existing_collections:
+            qdrant_client.create_collection(
+                collection_name=collection_name,
+                vectors_config=VectorParams(
+                    size=384,  # Ensure this matches embedding dimensions
+                    distance=Distance.COSINE
+                )
+            )
+            logging.info(f"Collection '{collection_name}' created.")
+        else:
+            logging.info(f"Collection '{collection_name}' already exists.")
+    except Exception as e:
+        logging.error(f" Error creating collection '{collection_name}': {e}")
+        raise
+
+def split_text_into_chunks(text, max_chunk_size=256):
+    """
+    Splits text into smaller, manageable chunks for indexing.
+    - Uses newline (`\n`) splitting if available.
+    - Falls back to `sent_tokenize()` if necessary.
+    - Splits large chunks further into smaller ones (max 256 tokens).
+    
+    Args:
+        text (str): Full document text.
+        max_chunk_size (int): Maximum token length per chunk.
+    
+    Returns:
+        list: List of properly split chunks.
+    """
+    # Try splitting by newlines if present
+    if "\n" in text:
+        chunks = [s.strip() for s in text.split("\n") if s.strip()]
+    else:
+        # Otherwise, use sentence tokenization
+        chunks = sent_tokenize(text)
+
+    # Ensure chunks are not too large (Break long sentences)
+    final_chunks = []
+    for chunk in chunks:
+        if len(chunk) > max_chunk_size:  
+            # Further split large chunks at punctuation
+            split_sub_chunks = re.split(r'(?<=[.?!])\s+', chunk)  # Split at sentence-ending punctuation
+            final_chunks.extend([s.strip() for s in split_sub_chunks if s.strip()])
+        else:
+            final_chunks.append(chunk)
+
+    logging.info(f" Split document into {len(final_chunks)} chunks.")
+    return final_chunks
+
+def index_document(collection_name, document_id, text, batch_size=100):
+    """
+    Indexes document text into Qdrant with improved chunking.
+    
+    Args:
+        collection_name (str): Name of the collection.
+        document_id (str): ID of the document.
+        text (str): Full document text.
+        batch_size (int): Number of chunks to process in a single batch.
+    
+    Returns:
+        dict: Status of the indexing operation.
+    """
+    try:
+        create_collection_if_not_exists(collection_name)
+
+        # 🔹 Improved chunking logic
+        chunks = split_text_into_chunks(text)
+
+        if not chunks:
+            logging.warning(" No valid chunks extracted for indexing.")
+            return {"status": "error", "message": "No valid chunks extracted"}
+
+        # 🔹 Process chunks in batches
+        for i in range(0, len(chunks), batch_size):
+            batch_chunks = chunks[i:i + batch_size]
+            embeddings = model.encode(batch_chunks).tolist()
+
+            points = []
+            for idx, (chunk, embedding) in enumerate(zip(batch_chunks, embeddings)):
+                chunk_id = str(uuid.uuid4())
+
+                payload = {
+                    "document_id": document_id,
+                    "text": chunk,
+                    "chunk_index": i + idx,
+                    "file_name": document_id  
+                }
+                points.append({
+                    "id": chunk_id,
+                    "vector": embedding,
+                    "payload": payload
+                })
+
+            # Upsert the batch into Qdrant
+            qdrant_client.upsert(collection_name=collection_name, points=points)
+            logging.info(f" Indexed batch {i // batch_size + 1} ({len(batch_chunks)} chunks).")
+
+        logging.info(f" Successfully indexed {len(chunks)} chunks for document '{document_id}'.")
+        return {"status": "success", "chunks": len(chunks)}
+
+    except Exception as e:
+        logging.error(f"Error indexing document '{document_id}': {e}")
+        return {"status": "error", "message": str(e)}

new.py

@@ -0,0 +1,14 @@
+from qdrant_client import QdrantClient
+import os
+
+# Connect to Qdrant
+client = QdrantClient(host="localhost", port=6333)
+
+# List all collections
+collections = client.get_collections()
+print("Available Collections:", collections)
+
+# Count indexed documents
+collection_name = "documents"  # Change if needed
+info = client.get_collection(collection_name)
+print("Collection Info:", info)

preprocess.py

@@ -0,0 +1,39 @@
+import re
+import nltk
+from nltk.corpus import stopwords
+from nltk.tokenize import word_tokenize
+from nltk.stem import WordNetLemmatizer
+import string  # Import the string module
+
+# Initialize lemmatizer and stopwords
+lemmatizer = WordNetLemmatizer()
+stop_words = set(stopwords.words('english'))
+
+# Text preprocessing function
+def preprocess_text(text):
+    # Convert text to lowercase
+    text = text.lower()
+
+    # Normalize line breaks and remove unnecessary spaces
+    text = re.sub(r'\s+', ' ', text.strip())
+
+    # Split alphanumeric combinations (e.g., "hello1234world" -> "hello 1234 world")
+    text = re.sub(r'([a-zA-Z]+)(\d+)', r'\1 \2', text)
+    text = re.sub(r'(\d+)([a-zA-Z]+)', r'\1 \2', text)
+
+    # Tokenize the text into words, numbers, and special characters
+    tokens = word_tokenize(text)
+
+    # Process tokens: lemmatize words, keep numbers and special characters
+    cleaned_tokens = []
+    for token in tokens:
+        if token.isalpha():  # Alphabetic words
+            if token not in stop_words:
+                cleaned_tokens.append(lemmatizer.lemmatize(token))
+        elif token.isnumeric():  # Numbers
+            cleaned_tokens.append(token)
+        elif not token.isalnum() and token not in string.punctuation:  # Special characters (excluding punctuation)
+            cleaned_tokens.append(token)
+
+    # Join the tokens back into a single string
+    return ' '.join(cleaned_tokens)

querying.py

@@ -0,0 +1,117 @@
+import os
+import torch
+import logging
+import preprocess
+from qdrant_client import QdrantClient
+from sentence_transformers import SentenceTransformer
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+# Configure Logging
+logging.basicConfig(level=logging.INFO)
+
+# Load Qdrant Configuration from Environment
+QDRANT_HOST = os.getenv("QDRANT_HOST", "localhost")
+QDRANT_PORT = int(os.getenv("QDRANT_PORT", 6333))
+
+# Initialize Qdrant Client
+qdrant_client = QdrantClient(host=QDRANT_HOST, port=QDRANT_PORT)
+
+# Load Sentence Transformer for Query Embeddings
+embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
+
+# Load GPT-2 from Hugging Face
+GPT2_MODEL_NAME = "gpt2"  # You can also use "gpt2-medium", "gpt2-large", "gpt2-xl" for larger versions
+tokenizer = AutoTokenizer.from_pretrained(GPT2_MODEL_NAME)
+gpt2_model = AutoModelForCausalLM.from_pretrained(
+    GPT2_MODEL_NAME,
+    torch_dtype=torch.float16,  # Lower memory usage
+    device_map="auto"  # Auto-select GPU if available
+)
+
+# Function to Generate Answer Using GPT-2
+def generate_answer(query, context):
+    """Generates a response using GPT-2 based on the retrieved context."""
+    if not context.strip():
+        return "I couldn't find relevant information."
+
+    prompt = f"""
+    Context: {context}
+
+    Question: {query}
+
+    Answer:
+    """
+
+    inputs = tokenizer(prompt, return_tensors="pt").to(gpt2_model.device)
+    outputs = gpt2_model.generate(
+        **inputs,
+        max_new_tokens=200,
+        temperature=0.7,
+        top_p=0.9
+    )
+    return tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+# Function to Query Documents from Qdrant
+def query_documents(collection_name, user_query, top_k=5, score_threshold=0.5):
+    """Queries Qdrant, retrieves matching documents, and generates an answer using GPT-2."""
+    try:
+        logging.info(f"🔍 Original Query: {user_query}")
+        processed_query = preprocess.preprocess_text(user_query)
+        logging.info(f" Preprocessed Query: {processed_query}")
+
+        # Generate Query Embedding
+        query_vector = embedding_model.encode(processed_query).tolist()
+
+        # Search in Qdrant
+        search_results = qdrant_client.search(
+            collection_name=collection_name,
+            query_vector=query_vector,
+            limit=top_k,
+            with_payload=True
+        )
+
+        if not search_results:
+            logging.warning(" No results found. Try increasing top_k or checking indexing.")
+
+        # Filter Results
+        filtered_results = [
+            {
+                "id": res.id,
+                "score": res.score,
+                "text": res.payload.get("text", ""),
+            }
+            for res in search_results if res.score >= score_threshold and "text" in res.payload
+        ]
+
+        # Extract Context for Answer Generation
+        context = " ".join(res["text"] for res in filtered_results) or "No relevant information found."
+        answer = generate_answer(user_query, context)
+
+        return {"answer": answer, "chunks": filtered_results}
+
+    except Exception as e:
+        logging.error(f"Error during query: {e}")
+        return {"error": str(e)}
+
+# Command-Line Execution
+if _name_ == "_main_":
+    import argparse
+
+    parser = argparse.ArgumentParser(description="Query documents with GPT-2")
+    parser.add_argument("--collection", type=str, default="documents", help="Qdrant collection name")
+    parser.add_argument("--query", type=str, required=True, help="Your search query")
+    parser.add_argument("--top-k", type=int, default=3, help="Number of results to return")
+    args = parser.parse_args()
+
+    logging.info(f"Querying for: '{args.query}'")
+    result = query_documents(args.collection, args.query, args.top_k)
+
+    if "error" in result:
+        logging.error(f" Error: {result['error']}")
+    else:
+        logging.info("\n===  Generated Answer ===")
+        print(result["answer"])
+
+        logging.info("\n===  Relevant Chunks ===")
+        for i, chunk in enumerate(result["chunks"]):
+            print(f"\nChunk {i+1} (Score: {chunk['score']:.3f}):\n{chunk['text']}")

requirements.txt

@@ -0,0 +1,6 @@
+flask
+python-docx
+PyPDF2
+sentence-transformers
+gradio
+nltk

tt.xml

@@ -0,0 +1,131 @@
+<mxfile host="app.diagrams.net" modified="2024-05-16T16:32:12.198Z" agent="Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36" version="24.5.3" etag="4Yj3hH6JYlqVxI2xUZbO" type="device">
+  <diagram name="Page-1" id="XIDM6lB2L0j4NQYgTgD7">
+    <mxGraphModel dx="1386" dy="778" grid="1" gridSize="10" guides="1" tooltips="1" connect="1" arrows="1" fold="1" page="1" pageScale="1" pageWidth="827" pageHeight="1169" math="0" shadow="0">
+      <root>
+        <mxCell id="0" />
+        <mxCell id="1" parent="0" />
+        <mxCell id="2" value="User/API" style="swimlane;fontStyle=1;align=center;verticalAlign=top;childLayout=stackLayout;horizontal=1;startSize=26;horizontalStack=0;resizeParent=1;resizeParentMax=0;resizeLast=0;collapsible=1;marginBottom=0;rounded=1;shadow=1;" vertex="1" parent="1">
+          <mxGeometry x="80" y="40" width="160" height="130" as="geometry" />
+        </mxCell>
+        <mxCell id="3" value="Uploads PDF" style="text;strokeColor=none;fillColor=none;align=left;verticalAlign=top;spacingLeft=4;spacingRight=4;overflow=hidden;rotatable=0;points=[[0,0.5],[1,0.5]];portConstraint=eastwest;" vertex="1" parent="2">
+          <mxGeometry y="26" width="160" height="26" as="geometry" />
+        </mxCell>
+        <mxCell id="4" value="Submits Query" style="text;strokeColor=none;fillColor=none;align=left;verticalAlign=top;spacingLeft=4;spacingRight=4;overflow=hidden;rotatable=0;points=[[0,0.5],[1,0.5]];portConstraint=eastwest;" vertex="1" parent="2">
+          <mxGeometry y="52" width="160" height="26" as="geometry" />
+        </mxCell>
+        <mxCell id="5" value="Receives Answer" style="text;strokeColor=none;fillColor=none;align=left;verticalAlign=top;spacingLeft=4;spacingRight=4;overflow=hidden;rotatable=0;points=[[0,0.5],[1,0.5]];portConstraint=eastwest;" vertex="1" parent="2">
+          <mxGeometry y="78" width="160" height="26" as="geometry" />
+        </mxCell>
+        <mxCell id="6" value="Document Loader (PyPDFLoader)" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#dae8fc;strokeColor=#6c8ebf;" vertex="1" parent="1">
+          <mxGeometry x="320" y="80" width="160" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="7" value="Text Splitter" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#d5e8d4;strokeColor=#82b366;" vertex="1" parent="1">
+          <mxGeometry x="560" y="80" width="160" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="8" value="Embedding Model (SentenceTransformers)" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#fff2cc;strokeColor=#d6b656;" vertex="1" parent="1">
+          <mxGeometry x="800" y="80" width="160" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="9" value="Qdrant DB" style="shape=cylinder3;whiteSpace=wrap;html=1;boundedLbl=1;backgroundOutline=1;size=15;fillColor=#f8cecc;strokeColor=#b85450;" vertex="1" parent="1">
+          <mxGeometry x="1040" y="80" width="160" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="10" value="Query Processor" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#e1d5e7;strokeColor=#9673a6;" vertex="1" parent="1">
+          <mxGeometry x="320" y="240" width="160" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="11" value="" style="endArrow=classic;html=1;exitX=1;exitY=0.5;exitDx=0;exitDy=0;entryX=0;entryY=0.5;entryDx=0;entryDy=0;" edge="1" parent="1" source="2" target="6">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint x="240" y="110" as="sourcePoint" />
+            <mxPoint x="320" y="110" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="12" value="Raw Text" style="edgeLabel;html=1;align=center;verticalAlign=middle;resizable=0;points=[];" vertex="1" connectable="0" parent="11">
+          <mxGeometry x="-0.2" y="-1" relative="1" as="geometry">
+            <mxPoint x="1" y="1" as="offset" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="13" value="" style="endArrow=classic;html=1;exitX=1;exitY=0.5;exitDx=0;exitDy=0;entryX=0;entryY=0.5;entryDx=0;entryDy=0;" edge="1" parent="1" source="6" target="7">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint x="480" y="110" as="sourcePoint" />
+            <mxPoint x="560" y="110" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="14" value="Splits into Chunks" style="edgeLabel;html=1;align=center;verticalAlign=middle;resizable=0;points=[];" vertex="1" connectable="0" parent="13">
+          <mxGeometry x="-0.2" y="-1" relative="1" as="geometry">
+            <mxPoint x="1" y="1" as="offset" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="15" value="" style="endArrow=classic;html=1;exitX=1;exitY=0.5;exitDx=0;exitDy=0;entryX=0;entryY=0.5;entryDx=0;entryDy=0;" edge="1" parent="1" source="7" target="8">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint x="720" y="110" as="sourcePoint" />
+            <mxPoint x="800" y="110" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="16" value="Generates Embeddings" style="edgeLabel;html=1;align=center;verticalAlign=middle;resizable=0;points=[];" vertex="1" connectable="0" parent="15">
+          <mxGeometry x="-0.2" y="-1" relative="1" as="geometry">
+            <mxPoint x="1" y="1" as="offset" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="17" value="" style="endArrow=classic;html=1;exitX=1;exitY=0.5;exitDx=0;exitDy=0;entryX=0;entryY=0.5;entryDx=0;entryDy=0;" edge="1" parent="1" source="8" target="9">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint x="960" y="110" as="sourcePoint" />
+            <mxPoint x="1040" y="120" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="18" value="Stores Vectors + Metadata" style="edgeLabel;html=1;align=center;verticalAlign=middle;resizable=0;points=[];" vertex="1" connectable="0" parent="17">
+          <mxGeometry x="-0.2" y="-1" relative="1" as="geometry">
+            <mxPoint x="1" y="1" as="offset" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="19" value="" style="endArrow=classic;html=1;exitX=0.5;exitY=1;exitDx=0;exitDy=0;entryX=0.5;entryY=0;entryDx=0;entryDy=0;" edge="1" parent="1" source="2" target="10">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint x="160" y="240" as="sourcePoint" />
+            <mxPoint x="400" y="240" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="20" value="Query Text" style="edgeLabel;html=1;align=center;verticalAlign=middle;resizable=0;points=[];" vertex="1" connectable="0" parent="19">
+          <mxGeometry x="-0.2" y="-1" relative="1" as="geometry">
+            <mxPoint x="1" y="1" as="offset" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="21" value="" style="endArrow=classic;html=1;exitX=1;exitY=0.5;exitDx=0;exitDy=0;entryX=0;entryY=0.5;entryDx=0;entryDy=0;" edge="1" parent="1" source="10" target="9">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint x="480" y="270" as="sourcePoint" />
+            <mxPoint x="1040" y="120" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="22" value="Vectorized Query" style="edgeLabel;html=1;align=center;verticalAlign=middle;resizable=0;points=[];" vertex="1" connectable="0" parent="21">
+          <mxGeometry x="-0.2" y="-1" relative="1" as="geometry">
+            <mxPoint x="1" y="1" as="offset" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="23" value="" style="endArrow=classic;html=1;exitX=0.5;exitY=0;exitDx=0;exitDy=0;entryX=0.5;entryY=1;entryDx=0;entryDy=0;" edge="1" parent="1" source="9" target="10">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint x="1120" y="160" as="sourcePoint" />
+            <mxPoint x="400" y="240" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="24" value="Top-K Chunks" style="edgeLabel;html=1;align=center;verticalAlign=middle;resizable=0;points=[];" vertex="1" connectable="0" parent="23">
+          <mxGeometry x="-0.2" y="-1" relative="1" as="geometry">
+            <mxPoint x="1" y="1" as="offset" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="25" value="" style="endArrow=classic;html=1;exitX=0.5;exitY=1;exitDx=0;exitDy=0;entryX=0.5;entryY=0;entryDx=0;entryDy=0;" edge="1" parent="1" source="10" target="2">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint x="400" y="300" as="sourcePoint" />
+            <mxPoint x="160" y="170" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="26" value="Answer" style="edgeLabel;html=1;align=center;verticalAlign=middle;resizable=0;points=[];" vertex="1" connectable="0" parent="25">
+          <mxGeometry x="-0.2" y="-1" relative="1" as="geometry">
+            <mxPoint x="1" y="1" as="offset" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="27" value="Port: 6333 (HTTP), 6334 (gRPC)" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" vertex="1" parent="1">
+          <mxGeometry x="1040" y="160" width="160" height="20" as="geometry" />
+        </mxCell>
+        <mxCell id="28" value="Metadata: source, page, chunk_index" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" vertex="1" parent="1">
+          <mxGeometry x="1040" y="180" width="160" height="20" as="geometry" />
+        </mxCell>
+      </root>
+    </mxGraphModel>
+  </diagram>
+</mxfile>