Update app.py

app.py

@@ -1,6 +1,8 @@
 # app.py
+
 import os
 import tempfile
+import hashlib
 import gradio as gr
 
 from embedding_manager import EmbeddingManager
@@ -10,21 +12,57 @@ from hedis_engine import HedisComplianceEngine
 APP_TITLE = "ChartWise AI"
 DEFAULT_MEASURE_YEAR = 2024
 
+# --- Simple in-process cache (per Space runtime) ---
+# Maps content_hash -> vectordb
+_VDB_CACHE = {}
+_LAST_HASH = None  # optional: for quick reuse/debug
+
+def _pdf_hash(pdf_bytes: bytes) -> str:
+    return hashlib.sha256(pdf_bytes).hexdigest()
+
+def _get_vectordb_from_bytes(pdf_bytes: bytes):
+    """
+    Return a cached vectordb for this PDF content if available,
+    otherwise build it once and cache it for subsequent calls.
+    """
+    global _VDB_CACHE, _LAST_HASH
+
+    content_hash = _pdf_hash(pdf_bytes)
+    _LAST_HASH = content_hash
+
+    if content_hash in _VDB_CACHE:
+        print(f"[CACHE] Using cached vectordb for hash {content_hash[:12]}...")
+        return _VDB_CACHE[content_hash]
+
+    # Not cached yet: create temp file, build embeddings once, then cache
+    with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp:
+        tmp.write(pdf_bytes)
+        pdf_path = tmp.name
+
+    try:
+        manager = EmbeddingManager(pdf_path)  # persists under ./embeddings/<temp_stem>
+        vectordb = manager.get_or_create_embeddings()
+        _VDB_CACHE[content_hash] = vectordb
+        print(f"[CACHE] Cached vectordb for hash {content_hash[:12]}.")
+        return vectordb
+    finally:
+        # Remove only the temp PDF; embeddings are persisted separately by EmbeddingManager
+        try:
+            os.unlink(pdf_path)
+        except Exception:
+            pass
+
 # --- Handlers ---
 def generate_patient_summary(pdf_file):
     try:
         if pdf_file is None:
             return "⚠️ Please upload a PDF file first."
-        with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp:
-            tmp.write(pdf_file)
-            pdf_path = tmp.name
 
-        manager = EmbeddingManager(pdf_path)
-        vectordb = manager.get_or_create_embeddings()
+        # Reuse embeddings if already built for this file
+        vectordb = _get_vectordb_from_bytes(pdf_file)
 
         summarizer = PatientChartSummarizer(vectordb)
         result = summarizer.summarize_chart()
-        os.unlink(pdf_path)
         return result
     except Exception as e:
         return f"❌ Error processing chart: {e}"
@@ -38,16 +76,11 @@ def generate_hedis_analysis(pdf_file, measure_name, measurement_year):
         if not measurement_year:
             return "⚠️ Please enter a measurement year."
 
-        with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp:
-            tmp.write(pdf_file)
-            pdf_path = tmp.name
-
-        manager = EmbeddingManager(pdf_path)
-        vectordb = manager.get_or_create_embeddings()
+        # Reuse embeddings if already built for this file
+        vectordb = _get_vectordb_from_bytes(pdf_file)
 
         hedis = HedisComplianceEngine(vectordb, measure_name, int(measurement_year))
         result = hedis.run()
-        os.unlink(pdf_path)
         return result
     except Exception as e:
         return f"❌ Error processing HEDIS analysis: {e}"
@@ -109,5 +142,4 @@ with gr.Blocks(theme=simple_theme, title="ChartWise AI") as interface:
 
 # Spaces auto-runs the script; these hints are fine, too:
 if __name__ == "__main__":
-    # On Spaces, port/host are managed, but these envs are also supported by Gradio.  [oai_citation:6‡Gradio](https://www.gradio.app/guides/environment-variables?utm_source=chatgpt.com)
     interface.queue().launch(server_name="0.0.0.0", server_port=int(os.environ.get("GRADIO_SERVER_PORT", "7860")))