app.py

import gradio as gr
import pandas as pd
import os
import time
from src.vector_db import UnifiedQdrant
from src.router import LearnedRouter
from src.comparison import ComparisonEngine
from config import COLLECTION_NAME, NUM_CLUSTERS, FRESHNESS_SHARD_ID, MRL_DIMS

# --- Initialization ---
print("Initializing dashVectorspace App...")

# 1. Initialize DB
db = UnifiedQdrant(
    collection_name=COLLECTION_NAME,
    vector_size=384,
    num_clusters=NUM_CLUSTERS,
    freshness_shard_id=FRESHNESS_SHARD_ID
)
db.initialize()

# 2. Initialize Router
ROUTER_PATH = "models/router_v1.pkl"
if os.path.exists(ROUTER_PATH):
    router = LearnedRouter.load(ROUTER_PATH)
else:
    print("WARNING: Router model not found. Creating a DUMMY router for demo UI.")
    router = LearnedRouter(model_type="lightgbm", n_clusters=NUM_CLUSTERS, mrl_dims=MRL_DIMS)
    router.predict = lambda x: (0, 0.99)

# 3. Initialize Engine
engine = ComparisonEngine(db, router, embedding_model_name="minilm")

# --- UI Logic ---
def run_comparison(query):
    if not query:
        return None, None, "Please enter a query."

    # Run Searches
    res_direct = engine.direct_search(query)
    res_xvector = engine.xvector_search(query)
    
    # --- 1. Benchmarking Table Data ---
    # Sketch Cols: Embedding Model | Router | dash Vector (Time, Shards) | Qdrant Search (Time, Shards)
    # We will format this as a Pandas DataFrame for the gr.Dataframe component
    
    df = pd.DataFrame({
        "Embedding Model": ["MiniLM-L6-v2"],
        "Router": ["LightGBM"],
        "dashVector (Optimized)": [f"{res_xvector['latency_ms']:.1f} ms | {res_xvector['shards_searched']} Shards"],
        "Qdrant (Baseline)": [f"{res_direct['latency_ms']:.1f} ms | {res_direct['shards_searched']} Shards"],
        "Savings": [f"{(1 - res_xvector['shards_searched']/res_direct['shards_searched'])*100:.1f}%"]
    })

    # --- 2. Search Results (Top 3) ---
    # Just showing top result text to prove it works, as per sketch focus on table
    def format_top_result(res_dict):
        if not res_dict["results"]:
            return "No results found."
        top_res = res_dict["results"][0]
        payload = top_res.payload
        text = payload.get("text", "No text") if payload else "No text"
        return f"Top Result: {text[:150]}..."

    results_preview = f"""
    <div style="display: flex; gap: 20px; margin-top: 10px;">
        <div style="flex: 1; padding: 10px; background: rgba(255,255,255,0.05); border-radius: 8px;">
            <strong>dashVector:</strong> {format_top_result(res_xvector)}
        </div>
        <div style="flex: 1; padding: 10px; background: rgba(255,255,255,0.05); border-radius: 8px;">
            <strong>Qdrant:</strong> {format_top_result(res_direct)}
        </div>
    </div>
    """
    
    return df, results_preview

# --- Custom CSS for Single Screen Layout ---
custom_css = """
body { background-color: #0b0f19; color: #e0e0e0; overflow: hidden; }
.gradio-container { max-width: 1200px !important; margin: 0 auto; height: 100vh; display: flex; flex-direction: column; justify-content: center; }
h1 { font-size: 2.5em; margin-bottom: 0.2em; text-align: center; background: -webkit-linear-gradient(45deg, #667eea, #764ba2); -webkit-background-clip: text; -webkit-text-fill-color: transparent; }
.input-box textarea { background: #1a1f2e !important; border: 1px solid #333 !important; font-size: 1.2em; }
.dataset-box { border: 1px solid #444; padding: 10px 20px; border-radius: 8px; text-align: center; font-weight: bold; background: #1a1f2e; display: inline-block; }
.scope-box { margin-top: 20px; padding: 15px; border-left: 4px solid #667eea; background: rgba(102, 126, 234, 0.1); }
.table-wrap { margin-top: 20px; }
footer { display: none !important; }
"""

# --- Gradio Layout ---
with gr.Blocks(title="dashVectorspace", theme=gr.themes.Soft(primary_hue="indigo", secondary_hue="slate"), css=custom_css) as demo:
    
    # Title
    gr.Markdown("# 🚀 dashVectorspace")
    
    # Search Section (Centered)
    with gr.Row(elem_id="search-row"):
        with gr.Column(scale=4):
            query_input = gr.Textbox(
                placeholder="Enter your search query here...", 
                show_label=False, 
                elem_classes="input-box",
                lines=1
            )
        with gr.Column(scale=1):
            submit_btn = gr.Button("Search", variant="primary", size="lg")

    # Benchmarking Table
    gr.Markdown("### ⚡ Benchmarking Results (Live)")
    results_table = gr.Dataframe(
        headers=["Embedding Model", "Router", "dashVector (Optimized)", "Qdrant (Baseline)", "Savings"],
        datatype=["str", "str", "str", "str", "str"],
        interactive=False,
        elem_classes="table-wrap"
    )
    
    # Result Preview (Hidden initially, shown after search)
    results_html = gr.HTML()

    # Footer Section: Dataset & Scope
    with gr.Row(style="margin-top: 40px; align-items: center;"):
        with gr.Column(scale=1):
            gr.HTML("""
            <div class="dataset-box">
                Dataset: MS MARCO
            </div>
            """)
        
        with gr.Column(scale=2):
            gr.HTML("""
            <div class="scope-box">
                <strong>Project Scope:</strong>
                <ul style="margin-top: 5px; padding-left: 20px;">
                    <li><strong>Learned Routing:</strong> Predicts target clusters to reduce search space by 90%.</li>
                    <li><strong>Custom Sharding:</strong> Explicit data partitioning for targeted retrieval.</li>
                    <li><strong>Matryoshka Embeddings:</strong> Adaptive dimensionality for high-speed filtering.</li>
                </ul>
            </div>
            """)

    # Event Listeners
    submit_btn.click(
        run_comparison, 
        inputs=[query_input], 
        outputs=[results_table, results_html]
    )
    query_input.submit(
        run_comparison, 
        inputs=[query_input], 
        outputs=[results_table, results_html]
    )

if __name__ == "__main__":
    demo.launch()