Update app.py

app.py

@@ -61,7 +61,11 @@ def extract_text_from_docx(docx_path):
     return text
 
 
-def process_and_query(text, file=None):
+def process_and_query(state, text, file=None):
+    # Initialize state on first run
+    if state is None:
+        state = {"processed_text": None, "conversation": []}
+
     # Check if a file is uploaded
     if file:
         # Get the uploaded file content
@@ -69,40 +73,24 @@ def process_and_query(text, file=None):
         if file.filename.endswith('.pdf'):
             with open("temp.pdf", "wb") as f:
                 f.write(content)
-            text = extract_text_from_pdf("temp.pdf")
+            state["processed_text"] = extract_text_from_pdf("temp.pdf")
         elif file.filename.endswith('.docx'):
             with open("temp.docx", "wb") as f:
                 f.write(content)
-            text = extract_text_from_docx("temp.docx")
+            state["processed_text"] = extract_text_from_docx("temp.docx")
         else:
             return {"error": "Unsupported file format"}
 
-    # Process the text and update FAISS index (similar to previous code)
-    sentences = text.split("\n")
-    embeddings = embedding_model.encode(sentences)
-    index.add(np.array(embeddings))
-
-    # Search the FAISS index
-    query_embedding = embedding_model.encode([text])
-    D, I = index.search(np.array(query_embedding), k=5)
-
-    top_documents = []
-    for idx in I[0]:
-        if idx != -1:  # Ensure that a valid index is found
-            top_documents.append(f"Document {idx}")
-
-    # Generate response using LLM (optional, commented out)
-    # You can replace this with your desired LLM interaction logic
-    # response = llm.run(inputs=text, max_length=100, temperature=0.7)["generated_text"]
-
-    return {"top_documents": top_documents, "response": None}  # Response from LLM (optional)
+    # Handle user question
+    if state["processed_text"] and text:
+        # Process the question and potentially use LLM for answering (optional)
+        question_embedding = embedding_model.encode([text])
+        # ... (logic to search the index and potentially use LLM for answering)
+        answer = "Answer retrieved from the document based on your question."  # Placeholder answer
 
+        # Update conversation history
+        state["conversation"].append({"question": text,
 
-# Define the Gradio interface
-interface = gr.Interface(
-    fn=process_and_query,
-    inputs={"text": gr.Textbox(label="Enter text or upload a file"),
-             "file": gr.FileUpload(label="Upload File (PDF or Word