Update app.py

app.py

@@ -53,7 +53,6 @@ def extract_priority_pages(file_path: str) -> str:
         with pdfplumber.open(file_path) as pdf:
             for i, page in enumerate(pdf.pages):
                 page_text = page.extract_text() or ""
-                # Include first 3 pages or pages with medical keywords
                 if i < 3 or any(re.search(rf'\b{kw}\b', page_text.lower()) for kw in MEDICAL_KEYWORDS):
                     text_chunks.append(f"=== Page {i+1} ===\n{page_text.strip()}")
         return "\n\n".join(text_chunks)
@@ -181,7 +180,7 @@ Analyze the medical records for clinical oversights. Provide a concise, evidence
 4. **Urgent Follow-up**:
    - Flag abnormal lab results, imaging, behaviors, or legal history needing immediate reassessment or referral.
 
-Medical Records (Chunk {0}):
+Medical Records (Chunk {0} of {1}):
 {{chunk}}
 
 Begin analysis:
@@ -191,9 +190,13 @@ Begin analysis:
                 if history and history[-1]["content"].startswith("⏳"):
                     history.pop()
 
-                # Process each chunk sequentially
+                # Process each chunk and stream results in real-time
                 for chunk_idx, chunk in enumerate(chunks, 1):
-                    prompt = prompt_template.format(chunk_idx, chunk=chunk)
+                    # Update UI with progress
+                    history.append({"role": "assistant", "content": f"🔄 Processing Chunk {chunk_idx} of {len(chunks)}..."})
+                    yield history, None
+
+                    prompt = prompt_template.format(chunk_idx, len(chunks), chunk=chunk)
                     chunk_response = ""
                     for chunk_output in agent.run_gradio_chat(
                         message=prompt,
@@ -212,15 +215,29 @@ Begin analysis:
                                     cleaned = clean_response(m.content)
                                     if cleaned:
                                         chunk_response += cleaned + "\n"
+                                        # Update UI with partial response
+                                        if history[-1]["content"].startswith("🔄"):
+                                            history[-1] = {"role": "assistant", "content": f"--- Analysis for Chunk {chunk_idx} ---\n{chunk_response.strip()}"}
+                                        else:
+                                            history[-1]["content"] = f"--- Analysis for Chunk {chunk_idx} ---\n{chunk_response.strip()}"
+                                        yield history, None
                         elif isinstance(chunk_output, str) and chunk_output.strip():
                             cleaned = clean_response(chunk_output)
                             if cleaned:
                                 chunk_response += cleaned + "\n"
+                                # Update UI with partial response
+                                if history[-1]["content"].startswith("🔄"):
+                                    history[-1] = {"role": "assistant", "content": f"--- Analysis for Chunk {chunk_idx} ---\n{chunk_response.strip()}"}
+                                else:
+                                    history[-1]["content"] = f"--- Analysis for Chunk {chunk_idx} ---\n{chunk_response.strip()}"
+                                yield history, None
+
+                    # Append completed chunk response to combined response
                     combined_response += f"--- Analysis for Chunk {chunk_idx} ---\n{chunk_response}\n"
 
-                # Update history with combined response
+                # Finalize UI with complete response
                 if combined_response:
-                    history.append({"role": "assistant", "content": combined_response.strip()})
+                    history[-1]["content"] = combined_response.strip()
                 else:
                     history.append({"role": "assistant", "content": "No oversights identified."})