Create APP

APP

@@ -0,0 +1,204 @@
+import gradio as gr
+import numpy as np
+import inspect
+import pandas as pd
+import torch
+import re
+from datasets import load_dataset
+from sentence_transformers import SentenceTransformer
+import faiss
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+
+# --- 1. LOAD GENERATIVE AI COMPONENT (Stabilized Flan-T5-Base) ---
+device = "cuda" if torch.cuda.is_available() else "cpu"
+print(f"Loading GenAI Component on {device}...")
+
+gen_tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-base")
+gen_model = AutoModelForSeq2SeqLM.from_pretrained("google/flan-t5-base").to(device)
+
+def generate_sales_pitch(user_query, company_name, sector, theme, description):
+    # Safer, simpler prompt to stop hallucinations
+    prompt = f"Write one short sentence explaining why {company_name} (a {theme} company) is a good investment for the topic '{user_query}'."
+
+    try:
+        inputs = gen_tokenizer(prompt, return_tensors="pt", max_length=128, truncation=True).to(device)
+
+        # Stabilized parameters
+        outputs = gen_model.generate(
+            **inputs,
+            max_new_tokens=40,
+            do_sample=False,
+            num_beams=4,
+            no_repeat_ngram_size=2,
+            early_stopping=True
+        )
+        pitch = gen_tokenizer.decode(outputs[0], skip_special_tokens=True)
+        return pitch
+    except Exception:
+        return f"{company_name} is a leading industry player in {theme}, aligning with '{user_query}'."
+
+
+# --- 2. AUTOMATIC SAFETY AUTO-LOADER (Dataset & Embeddings) ---
+if 'df' not in globals():
+    print("Auto-loading dataset 'Yoel125/synthetic-companies-12k' from Hugging Face...")
+    df = pd.DataFrame(load_dataset('Yoel125/synthetic-companies-12k', split='train'))
+    df['full_text'] = df['sector'] + " - " + df['theme'] + ": " + df['description']
+
+if 'embedding_model' not in globals():
+    print("Auto-loading embedding model 'paraphrase-MiniLM-L3-v2'...")
+    embedding_model = SentenceTransformer('paraphrase-MiniLM-L3-v2')
+
+if 'faiss_index' not in globals():
+    try:
+        print("Loading saved embeddings from company_embeddings.npy...")
+        embeddings = np.load('company_embeddings.npy')
+        faiss_index = faiss.IndexFlatL2(embeddings.shape[1])
+        faiss_index.add(np.array(embeddings).astype('float32'))
+    except FileNotFoundError:
+        print("Saved embeddings not scratch...")
+        embeddings = embedding_model.encode(df['full_text'].tolist(), show_progress_bar=False)
+        faiss_index = faiss.IndexFlatL2(embeddings.shape[1])
+        faiss_index.add(np.array(embeddings).astype('float32'))
+
+
+# --- 3. RECOMMENDATION ENGINE LOGIC ---
+all_sectors = ["All Sectors"] + sorted(list(df['sector'].unique()))
+
+def recommend_investment(user_query, selected_sector, top_k=3):
+    # 1. Check if completely empty
+    if (not user_query or not str(user_query).strip()) and selected_sector == "All Sectors":
+        yield "Please enter an investment thesis or keyword in the text box above, or select a specific industry sector from the dropdown menu."
+        return
+
+    # 2. English Language Check (Blocks Hebrew, Arabic, etc.)
+    if user_query and str(user_query).strip():
+        non_english_chars = sum(1 for char in str(user_query) if ord(char) > 127)
+        if non_english_chars > 2:
+            yield "⚠️ **Language Not Supported:** SectorSync AI is currently optimized exclusively for English data. Please write your investment thesis in English and try again."
+            return
+
+    # 3. Auto-fill sector if text box is empty
+    if not user_query or not str(user_query).strip():
+        user_query = f"innovative {selected_sector} companies"
+
+    yield "🔍 Searching for matching companies and generating AI insights... this can take a few seconds."
+
+    # 4. Search Execution
+    if selected_sector != "All Sectors":
+        enriched_query = f"{selected_sector} industry B2B company specializing in: {user_query}"
+    else:
+        enriched_query = f"B2B investment opportunity specializing in: {user_query}"
+
+    query_vector = embedding_model.encode([enriched_query])
+    search_k = 25 if selected_sector != "All Sectors" else top_k
+    distances, indices = faiss_index.search(np.array(query_vector).astype('float32'), k=search_k)
+
+    matches = []
+    for i in range(search_k):
+        idx = indices[0][i]
+        sim_score = 1 / (1 + distances[0][i])
+        company_row = df.iloc[idx]
+        if selected_sector != "All Sectors" and company_row['sector'] != selected_sector:
+            continue
+        matches.append((company_row, sim_score))
+        if len(matches) == top_k:
+            break
+
+    if not matches:
+        for i in range(min(top_k, len(indices[0]))):
+            idx = indices[0][i]
+            sim_score = 1 / (1 + distances[0][i])
+            matches.append((df.iloc[idx], sim_score))
+
+    output_markdown = f"### Top {len(matches)} AI-Recommended Matches for: *'{user_query}'*\n\n"
+    if selected_sector != "All Sectors":
+        output_markdown += f"**Filtered by Sector:** `{selected_sector}`\n\n---\n\n"
+    else:
+        output_markdown += "---\n\n"
+
+    match_labels = ["🥇 Strongest Match", "🥈 Close Match", "🥉 Close Match"]
+
+    for rank, (row, score) in enumerate(matches, 1):
+        c_name = row['company_name']
+        ticker = row.get('ticker', 'N/A')
+        sector = row['sector']
+        theme = row['theme']
+        desc = row['description']
+
+        try:
+            pitch = generate_sales_pitch(user_query, c_name, sector, theme, desc)
+        except Exception:
+            pitch = f"An exceptional strategic match for {user_query} within the {sector} space."
+
+        tier_label = match_labels[rank - 1] if rank <= len(match_labels) else "Match"
+
+        # The Fix: No bullet points, reads like a clean, professional report!
+        output_markdown += f"#### #{rank}. {c_name} (`{ticker}`) — *{sector}*\n\n"
+        output_markdown += f"**{tier_label}** (Similarity Score: `{score*100:.1f}%`)\n\n"
+        output_markdown += f"**Industry:** {theme}\n\n"
+        output_markdown += f"**GenAI Investment Pitch:** *'{pitch}'*\n\n"
+        output_markdown += f"**Company Overview:** {desc[:200]}...\n\n---\n\n"
+
+        # Streams the results instantly!
+        yield output_markdown
+
+
+# --- 4. GRADIO USER INTERFACE ---
+custom_css = """
+body, .gradio-container {
+    background-color: #121212 !important;
+    color: #f0f4f8 !important;
+}
+.markdown-text, h1, h2, h3, h4, p, li, span, label {
+    color: #f0f4f8 !important;
+}
+button.primary {
+    background-color: #10b981 !important;
+    border-color: #10b981 !important;
+    color: #121212 !important;
+    font-weight: bold !important;
+}
+button.primary:hover {
+    background-color: #059669 !important;
+}
+"""
+
+with gr.Blocks(theme=gr.themes.Base(), css=custom_css, title="SectorSync AI") as demo:
+    gr.Markdown("# SectorSync AI")
+    gr.Markdown("Cut Through the Market Noise — Find Your Next Winning Stock in Seconds. Discover high-growth companies matching your investment thesis using FAISS Vector Search and Generative AI.")
+    gr.Markdown("*Similarity Score reflects how closely a company profile matches your query in AI-embedding space — a relative ranking signal, not a calibrated financial confidence rating.*")
+
+    with gr.Row():
+        with gr.Column(scale=2):
+            query_input = gr.Textbox(
+                label="What is your investment thesis or topic?",
+                placeholder="e.g., autonomous robotics, clean battery storage, gene therapy for rare diseases...",
+                lines=2
+            )
+        with gr.Column(scale=1):
+            sector_dropdown = gr.Dropdown(
+                choices=all_sectors,
+                value="All Sectors",
+                label="Sector"
+            )
+
+    search_button = gr.Button("Find Investment Matches", variant="primary")
+
+    gr.Examples(
+        examples=[
+            ["Artificial Intelligence and Machine Learning in Healthcare", "Healthcare"],
+            ["Next-generation Renewable Energy and Solar Battery Storage", "Energy"],
+            ["Autonomous Robotics and Supply Chain Logistics Automation", "Industrials"]
+        ],
+        inputs=[query_input, sector_dropdown],
+        label="Quick Starters (1-Click Example Searches)"
+    )
+
+    results_output = gr.Markdown(label="Recommendation Results")
+
+    search_button.click(fn=recommend_investment, inputs=[query_input, sector_dropdown], outputs=results_output)
+    query_input.submit(fn=recommend_investment, inputs=[query_input, sector_dropdown], outputs=results_output)
+
+if __name__ == "__main__":
+    print("Launching SectorSync AI Recommender App...")
+    demo.launch(share=True)