Create app.py

app.py

@@ -0,0 +1,132 @@
+import os
+import sys
+
+# --- HUGGING FACE CHROMADB FIX ---
+# This is critical for deployment on HF Spaces to prevent SQLite errors
+try:
+    __import__('pysqlite3')
+    sys.modules['sqlite3'] = sys.modules.pop('pysqlite3')
+except ImportError:
+    pass
+# ---------------------------------
+
+import gradio as gr
+from langchain_openai import ChatOpenAI, OpenAIEmbeddings
+from langchain_community.document_loaders import PyPDFLoader
+from langchain_text_splitters import RecursiveCharacterTextSplitter
+from langchain_chroma import Chroma
+from langchain.chains import RetrievalQA
+from langchain.prompts import PromptTemplate
+
+# Global variables to store state (sufficient for demo purposes)
+vectorstore = None
+qa_chain = None
+
+def process_pdf(file_path, api_key):
+    global vectorstore, qa_chain
+    
+    if not api_key:
+        return "⚠️ Error: Please enter your OpenAI API Key first."
+    
+    os.environ["OPENAI_API_KEY"] = api_key
+    
+    try:
+        # 1. Load PDF
+        loader = PyPDFLoader(file_path)
+        docs = loader.load()
+        
+        # 2. Split Text
+        text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
+        splits = text_splitter.split_documents(docs)
+        
+        # 3. Embed & Store (Chroma)
+        embeddings = OpenAIEmbeddings()
+        vectorstore = Chroma.from_documents(documents=splits, embedding=embeddings)
+        
+        # 4. Create Retrieval Chain with "Auditor" Persona
+        llm = ChatOpenAI(model_name="gpt-4o", temperature=0)
+        
+        # Custom Prompt to enforce "Audit" style behavior
+        audit_template = """You are Veritas, an AI Compliance Auditor. 
+        Use the following pieces of context to answer the question at the end. 
+        
+        RULES:
+        1. If the answer is in the text, state it clearly and reference the context.
+        2. If the answer is NOT in the text, you must explicitly state: "FAIL: Information not found in source document."
+        3. Do not hallucinate or guess. 
+        
+        Context: {context}
+        
+        Question: {question}
+        
+        Verdict:"""
+        
+        QA_CHAIN_PROMPT = PromptTemplate.from_template(audit_template)
+        
+        qa_chain = RetrievalQA.from_chain_type(
+            llm=llm,
+            chain_type="stuff",
+            retriever=vectorstore.as_retriever(),
+            chain_type_kwargs={"prompt": QA_CHAIN_PROMPT}
+        )
+        
+        return "✅ Document Processed Successfully. The Veritas Auditor is ready."
+        
+    except Exception as e:
+        return f"❌ Error processing document: {str(e)}"
+
+def audit_query(query):
+    global qa_chain
+    if not qa_chain:
+        return "⚠️ Please upload a document first."
+    
+    try:
+        response = qa_chain.invoke(query)
+        return response['result']
+    except Exception as e:
+        return f"Error: {str(e)}"
+
+# --- GRADIO INTERFACE ---
+with gr.Blocks(theme=gr.themes.Soft(primary_hue="slate")) as demo:
+    
+    gr.Markdown(
+        """
+        # 🛡️ Veritas: AI Compliance Auditor
+        ### Automated RAG Hallucination Detection for Financial Documentation
+        """
+    )
+    
+    with gr.Row():
+        with gr.Column(scale=1):
+            api_input = gr.Textbox(
+                label="OpenAI API Key", 
+                type="password", 
+                placeholder="sk-..."
+            )
+            file_input = gr.File(
+                label="Upload Financial Report (PDF)", 
+                file_types=[".pdf"]
+            )
+            upload_btn = gr.Button("Initialize Auditor", variant="primary")
+            status_output = gr.Textbox(label="System Status", interactive=False)
+            
+        with gr.Column(scale=2):
+            query_input = gr.Textbox(label="Audit Query")
+            audit_btn = gr.Button("Run Audit Check")
+            response_output = gr.Textbox(label="Auditor Verdict", lines=10)
+
+    # Button Actions
+    upload_btn.click(
+        process_pdf, 
+        inputs=[file_input, api_input], 
+        outputs=status_output
+    )
+    
+    audit_btn.click(
+        audit_query, 
+        inputs=query_input, 
+        outputs=response_output
+    )
+
+if __name__ == "__main__":
+    demo.launch()