Create app.py

(The "Orchestrator". Connects the UI -> Tools -> Qwen.)

app.py

@@ -0,0 +1,114 @@
+# app.py
+import gradio as gr
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
+from threading import Thread
+
+# Import our custom modules
+import guardrails
+import tools_processing
+
+# --- 1. CONFIGURATION ---
+MODEL_ID = "Qwen/Qwen2.5-1.5B-Instruct" # The Brain
+
+print(">>> INITIALIZING ORCHESTRATOR...")
+print(">>> LOADING QWEN (This takes 1 min)...")
+
+try:
+    tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+    model = AutoModelForCausalLM.from_pretrained(
+        MODEL_ID, 
+        torch_dtype=torch.float32, 
+        device_map="cpu"
+    )
+    print(">>> QWEN LOADED. SYSTEMS ONLINE.")
+except Exception as e:
+    print(f"CRITICAL ERROR: {e}")
+
+# --- 2. THE ORCHESTRATOR LOGIC ---
+def run_orchestrator(message, history, file_upload, url_input):
+    """
+    The Central Nervous System.
+    1. Checks for inputs (Files vs Links).
+    2. Runs Tools (OCR / Scraper).
+    3. Builds Context.
+    4. Streams Qwen Response.
+    """
+    
+    # A. Run Tools
+    extracted_invoice_text = ""
+    external_context = ""
+    
+    # Tool: Web Scraper
+    if url_input:
+        external_context = tools_processing.scrape_public_link(url_input)
+    
+    # Tool: OCR
+    if file_upload:
+        # Note: In a chat flow, we might have processed this already. 
+        # For simplicity, we re-process or check state.
+        _, extracted_invoice_text = tools_processing.perform_ocr(file_upload)
+    
+    # B. Build Prompt with Guardrails
+    # If we have file text, we inject it. If not, we assume user is just chatting.
+    system_block = guardrails.get_guardrails(external_context, extracted_invoice_text)
+    
+    # C. Prepare LLM Input
+    # Combine System Prompt + Chat History + Latest Message
+    full_conversation = f"{system_block}\n"
+    
+    for user_msg, bot_msg in history:
+        full_conversation += f"<|im_start|>user\n{user_msg}<|im_end|>\n<|im_start|>assistant\n{bot_msg}<|im_end|>\n"
+    
+    full_conversation += f"<|im_start|>user\n{message}<|im_end|>\n<|im_start|>assistant\n"
+    
+    # D. Stream Generation (Low Latency)
+    inputs = tokenizer(full_conversation, return_tensors="pt")
+    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
+    
+    generation_kwargs = dict(
+        inputs, 
+        streamer=streamer, 
+        max_new_tokens=512, 
+        temperature=0.7
+    )
+    
+    thread = Thread(target=model.generate, kwargs=generation_kwargs)
+    thread.start()
+    
+    partial_response = ""
+    for new_text in streamer:
+        partial_response += new_text
+        yield partial_response
+
+# --- 3. UI SETUP ---
+def process_file_display(file):
+    """Helper to show the image after upload"""
+    img, _ = tools_processing.perform_ocr(file)
+    return img
+
+with gr.Blocks(theme=gr.themes.Soft(), title="MCP Invoice Orchestrator") as demo:
+    gr.Markdown("## 🧠 MCP Invoice Orchestrator (Powered by Qwen)")
+    gr.Markdown("Upload an invoice + Add a Notion Link. The Agent will enforce your rules.")
+    
+    with gr.Row():
+        # Left: Tools
+        with gr.Column(scale=1):
+            file_in = gr.File(label="📄 Invoice (PDF/Img)")
+            img_preview = gr.Image(label="Scan Preview", height=250)
+            url_in = gr.Textbox(label="🔗 Notion Public Link (Context)", placeholder="https://notion.so/public-page...")
+            
+            # Update preview on upload
+            file_in.upload(process_file_display, inputs=file_in, outputs=img_preview)
+            
+        # Right: Chat Interface
+        with gr.Column(scale=2):
+            chatbot = gr.ChatInterface(
+                fn=run_orchestrator,
+                additional_inputs=[file_in, url_in],
+                examples=[[x["query"]] for x in guardrails.CHAT_STARTERS],
+                title="Agent Interaction"
+            )
+
+if __name__ == "__main__":
+    demo.launch()