Add application file

app.py

@@ -52,17 +52,18 @@ curated_qa_pairs = [
     }
 ]
 
-def get_curated_context(query, curated_qa, embed_model):
+def get_curated_pair(query, curated_qa, embed_model, threshold=1.0):
     """
     Retrieve the most relevant curated Q/A pair based on the user's query.
-    Returns a formatted string if the similarity (using L2 distance) is below a threshold.
+    Returns the QA dictionary if the similarity (using L2 distance) is below the threshold,
+    otherwise returns None.
     """
     curated_questions = [qa["question"] for qa in curated_qa]
     query_embedding = embed_model.encode([query]).astype('float32')
     curated_embeddings = embed_model.encode(curated_questions, show_progress_bar=False)
     curated_embeddings = np.array(curated_embeddings).astype('float32')
     
-    # Build a temporary FAISS index for curated questions
+    # Build a temporary FAISS index for the curated questions
     dimension = curated_embeddings.shape[1]
     curated_index = faiss.IndexFlatL2(dimension)
     curated_index.add(curated_embeddings)
@@ -70,15 +71,11 @@ def get_curated_context(query, curated_qa, embed_model):
     k = 1
     distances, indices = curated_index.search(query_embedding, k)
     
-    # Define a threshold for relevance (tune this value as needed)
-    threshold = 1.0
     if distances[0][0] < threshold:
         idx = indices[0][0]
-        qa_pair = curated_qa[idx]
-        # Return a formatted string with the curated question and answer.
-        return f"Curated Q/A Pair:\nQuestion: {qa_pair['question']}\nAnswer: {qa_pair['answer']}\n"
+        return curated_qa[idx]
     else:
-        return ""
+        return None
 
 def main():
     st.title("PDF Question-Answering App")
@@ -108,23 +105,27 @@ def main():
         for idx in indices[0]:
             pdf_context += chunks[idx] + "\n"
         
-        # Get curated Q/A context if available
-        curated_context = get_curated_context(query, curated_qa_pairs, embed_model)
-        
         base_context = st.session_state.conversation_history + "\n"
         
-        if curated_context:
+        # Check for a curated Q/A pair
+        curated_pair = get_curated_pair(query, curated_qa_pairs, embed_model)
+        
+        if curated_pair:
             st.write("A curated Q/A pair was found and will be used for the answer by default.")
-            # Provide an option to override with full PDF context
+            # Option to override with full PDF context
             use_full_data = st.checkbox("Check Full Data", value=False)
-            if use_full_data:
-                context_to_use = base_context + pdf_context
+            if not use_full_data:
+                # Directly display the curated answer without running the QA pipeline
+                answer = curated_pair["answer"]
+                st.write(answer)
+                st.session_state.conversation_history += f"AI: {answer}\n"
+                return  # Exit the function after displaying the curated answer
             else:
-                context_to_use = base_context + curated_context
+                context_to_use = base_context + pdf_context
         else:
             context_to_use = base_context + pdf_context
         
-        # Optionally, you can also provide an expander to show the full PDF context.
+        # Provide an expander to show the full PDF context if desired
         with st.expander("Show Full PDF Context"):
             st.write(pdf_context)