Update back.branched.PDFAddedRAG.03282025.app.py

back.branched.PDFAddedRAG.03282025.app.py

@@ -101,9 +101,10 @@ def generate_filename(prompt, file_type, original_name=None):
     safe_date_time = datetime.now(central).strftime("%m%d_%H%M")
     if original_name and file_type == "md":  # For images
         base_name = os.path.splitext(original_name)[0]
-        safe_prompt = re.sub(r'[<>:"/\\|?*\n]', ' ', prompt).strip()[:100]
-        return f"{safe_date_time}_{safe_prompt}_{base_name}.{file_type}"
-    safe_prompt = re.sub(r'[<>:"/\\|?*\n]', ' ', prompt).strip()[:240]
+        safe_prompt = re.sub(r'[<>:"/\\|?*\n]', ' ', prompt).strip()[:50]
+        file_stem = f"{safe_date_time}_{safe_prompt}_{base_name}"[:100]  # Cap at 100 chars
+        return f"{file_stem}.{file_type}"
+    safe_prompt = re.sub(r'[<>:"/\\|?*\n]', ' ', prompt).strip()[:100]  # Cap at 100 chars
     return f"{safe_date_time}_{safe_prompt}.{file_type}"
 
 def create_and_save_file(content, file_type="md", prompt=None, original_name=None, should_save=True):
@@ -308,7 +309,7 @@ def generate_questions(pdf_path):
         pdf = PdfReader(f)
         for page in pdf.pages:
             text += page.extract_text() or ""
-    prompt = f"Can you generate a question that can only be answered from this document?:\n{text[:2000]}\n\n"
+    prompt = f"Generate a 10-question quiz with answers based only on this document. Format as markdown with numbered questions and answers:\n{text[:2000]}\n\n"
     response = client.chat.completions.create(
         model="gpt-4o-2024-05-13",
         messages=[{"role": "user", "content": prompt}]
@@ -320,7 +321,12 @@ def process_rag_query(query, vector_store_id):
         response = client.chat.completions.create(
             model="gpt-4o-2024-05-13",
             messages=[{"role": "user", "content": query}],
-            tools=[{"type": "file_search", "file_search": {"vector_store_ids": [vector_store_id]}}],
+            tools=[{
+                "type": "file_search",
+                "file_search": {
+                    "vector_store_ids": [vector_store_id]
+                }
+            }],
             tool_choice="auto"
         )
         tool_calls = response.choices[0].message.tool_calls if response.choices[0].message.tool_calls else []
@@ -331,26 +337,28 @@ def process_rag_query(query, vector_store_id):
 
 def evaluate_rag(vector_store_id, questions_dict):
     k = 5
-    total_queries = len(questions_dict)
+    total_queries = len(questions_dict) * 10  # 10 questions per PDF
     correct_retrievals_at_k = 0
     reciprocal_ranks = []
     average_precisions = []
     
-    for filename, query in questions_dict.items():
-        expected_file = filename
-        response, tool_calls = process_rag_query(query, vector_store_id)
-        if not tool_calls:
-            continue
-        retrieved_files = [call.arguments.get("file_id", "") for call in tool_calls if "file_search" in call.type][:k]
-        if expected_file in retrieved_files:
-            rank = retrieved_files.index(expected_file) + 1
-            correct_retrievals_at_k += 1
-            reciprocal_ranks.append(1 / rank)
-            precisions = [1 if f == expected_file else 0 for f in retrieved_files[:rank]]
-            average_precisions.append(sum(precisions) / len(precisions))
-        else:
-            reciprocal_ranks.append(0)
-            average_precisions.append(0)
+    for filename, quiz in questions_dict.items():
+        questions = re.findall(r"\d+\.\s(.*?)\n\s*Answer:\s(.*?)\n", quiz, re.DOTALL)
+        for question, _ in questions:
+            expected_file = filename
+            response, tool_calls = process_rag_query(question, vector_store_id)
+            if not tool_calls:
+                continue
+            retrieved_files = [call.arguments.get("file_id", "") for call in tool_calls if "file_search" in call.type][:k]
+            if expected_file in retrieved_files:
+                rank = retrieved_files.index(expected_file) + 1
+                correct_retrievals_at_k += 1
+                reciprocal_ranks.append(1 / rank)
+                precisions = [1 if f == expected_file else 0 for f in retrieved_files[:rank]]
+                average_precisions.append(sum(precisions) / len(precisions))
+            else:
+                reciprocal_ranks.append(0)
+                average_precisions.append(0)
     
     recall_at_k = correct_retrievals_at_k / total_queries if total_queries else 0
     mrr = sum(reciprocal_ranks) / total_queries if total_queries else 0
@@ -367,12 +375,26 @@ def rag_pdf_gallery():
             stats = upload_pdf_files_to_vector_store(vector_store_details["id"], pdf_paths)
             st.json(stats)
         
-        with st.spinner("Generating evaluation questions..."):
+        col1, col2, col3 = st.columns(3)
+        with col1:
+            if st.button("📝 Quiz"):
+                st.session_state["rag_prompt"] = "Generate a 10-question quiz with answers based only on this document."
+        with col2:
+            if st.button("📑 Summary"):
+                st.session_state["rag_prompt"] = "Summarize this per page and output as markdown outline with emojis and numbered outline with multiple levels summarizing everything unique per page in method steps or fact steps."
+        with col3:
+            if st.button("🔍 Key Facts"):
+                st.session_state["rag_prompt"] = "Extract 10 key facts from this document in markdown with emojis."
+        
+        with st.spinner("Generating questions..."):
             questions_dict = {os.path.basename(p): generate_questions(p) for p in pdf_paths}
-            st.json(questions_dict)
+            st.markdown("### Generated Quiz")
+            for filename, quiz in questions_dict.items():
+                st.markdown(f"#### {filename}")
+                st.markdown(quiz)
         
-        query = st.text_input("Ask a question about the PDFs:")
-        if query:
+        query = st.text_input("Ask a question about the PDFs:", value=st.session_state.get("rag_prompt", ""))
+        if query and st.button("Submit RAG Query"):
             with st.spinner("Processing RAG query..."):
                 response, tool_calls = process_rag_query(query, vector_store_details["id"])
                 if response:
@@ -381,6 +403,7 @@ def rag_pdf_gallery():
                     for call in tool_calls:
                         if "file_search" in call.type:
                             st.json(call.arguments)
+            st.rerun()
         
         if st.button("Evaluate RAG Performance"):
             with st.spinner("Evaluating..."):
@@ -465,7 +488,7 @@ def main():
     option = st.selectbox("Select Input Type", ("Text", "Image", "Audio", "Video", "ArXiv Search", "RAG PDF Gallery"))
     
     if option == "Text":
-        default_text = "emojis in markdown. Maybe a buckeyball feature rating comparing them against each other in markdown emoji outline or tables."
+        default_text = "Create a summary of PDF py libraries and usage in py with emojis in markdown. Maybe a buckeyball feature rating comparing them against each other in markdown emoji outline or tables."
         col1, col2 = st.columns([1, 5])
         with col1:
             if st.button("📝 MD", key="md_button"):
@@ -475,9 +498,10 @@ def main():
                 st.rerun()
         with col2:
             text_input = st.text_input("Enter your text:", value=st.session_state.get("text_input", ""), key="text_input_field")
-        if text_input and text_input != st.session_state.get("text_input", ""):  # Only process if changed
+        if text_input and st.button("Submit Text"):
             with st.spinner("Processing..."):
                 process_text(text_input)
+            st.rerun()
     
     elif option == "Image":
         col1, col2 = st.columns(2)
@@ -489,7 +513,7 @@ def main():
                 st.session_state["image_prompt"] = "Show electronic text of text in the image."
         text_input = st.text_input("Image Prompt:", value=st.session_state.get("image_prompt", "Describe this image and list ten facts in a markdown outline with emojis."))
         image_input = st.file_uploader("Upload an image (max 200MB)", type=["png", "jpg", "jpeg"], accept_multiple_files=False)
-        if image_input and text_input:
+        if image_input and text_input and st.button("Submit Image"):
             if image_input.size > 200 * 1024 * 1024:
                 st.error("Image exceeds 200MB limit.")
             else:
@@ -503,13 +527,13 @@ def main():
         text_input = st.text_input("Audio Prompt:", value="Summarize this audio transcription in Markdown.")
         audio_input = st.file_uploader("Upload an audio file (max 200MB)", type=["mp3", "wav", "flac", "m4a"], accept_multiple_files=False)
         audio_bytes = audio_recorder()
-        if audio_bytes:
+        if audio_bytes and text_input and st.button("Submit Audio Recording"):
             with open("recorded_audio.wav", "wb") as f:
                 f.write(audio_bytes)
             with st.spinner("Processing..."):
                 process_audio(audio_bytes, text_input)
             st.rerun()
-        elif audio_input and text_input:
+        elif audio_input and text_input and st.button("Submit Audio File"):
             with st.spinner("Processing..."):
                 process_audio(audio_input, text_input)
             st.rerun()
@@ -517,7 +541,7 @@ def main():
     elif option == "Video":
         text_input = st.text_input("Video Prompt:", value="Summarize this video and its transcription in Markdown.")
         video_input = st.file_uploader("Upload a video file (max 200MB)", type=["mp4"], accept_multiple_files=False)
-        if video_input and text_input:
+        if video_input and text_input and st.button("Submit Video"):
             if video_input.size > 200 * 1024 * 1024:
                 st.error("Video exceeds 200MB limit.")
             else:
@@ -527,7 +551,7 @@ def main():
     
     elif option == "ArXiv Search":
         query = st.text_input("AI Search ArXiv Scholarly Articles:")
-        if query:
+        if query and st.button("Search ArXiv"):
             with st.spinner("Searching ArXiv..."):
                 result = search_arxiv(query)
                 st.markdown(result)