Update app.py

app.py

@@ -13,7 +13,7 @@ from bloatectomy import bloatectomy
 from werkzeug.utils import secure_filename
 from langchain_groq import ChatGroq
 from typing_extensions import TypedDict, NotRequired
-from werkzeug.utils import secure_filename
+
 # --- Logging ---
 logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] %(message)s")
 logger = logging.getLogger("patient-assistant")
@@ -57,18 +57,17 @@ def clean_notes_with_bloatectomy(text: str, style: str = "remov") -> str:
         return text
 
 # --- Agent prompt instructions ---
+# Important change: the assistant should NOT insist on PID at the start of conversation.
+# It should only ask for patient ID when the user asks for something that requires accessing previous medical records.
+# Also, avoid asking for additional identity details (name/DOB) purely for "verification" — accept PID as sufficient for record access in this flow.
 PATIENT_ASSISTANT_PROMPT = """
-You are a patient assistant helping to analyze medical records and reports. Your primary task is to get the patient ID (PID) from the user at the start of the conversation.
+You are a patient assistant helping to analyze medical records and reports. You should be helpful for general medical questions even when no patient ID (PID) is provided.
 
-Once you have the PID, you will be provided with a summary of the patient's medical reports. Use this information, along with the conversation history, to provide a comprehensive response.
-
-Your tasks include:
-- **First, ask for the patient ID.** Do not proceed with any other task until you have the PID.
-- Analyzing medical records and reports to detect anomalies, redundant tests, or misleading treatments.
-- Suggesting preventive care based on the overall patient health history.
-- Optimizing healthcare costs by comparing past visits and treatments.
-- Offering personalized lifestyle recommendations.
-- Generating a natural, helpful reply to the user.
+Behavior rules (follow these strictly):
+- Do NOT ask for the patient ID at the start of every conversation. Only request the PID when the user's question explicitly requires accessing prior medical records (for example: "show my previous lab report", "what does my thyroid test from last month say", "what was my doctor's note for PID 12345", etc.).
+- When you do ask for a PID, be concise and ask only for the PID (e.g., "Please provide the patient ID (PID) to retrieve previous records."). Do not request name/DOB/other verification unless the user explicitly asks for an extra verification step.
+- If the user supplies a PID in their message (patterns like "pid 5678", "p5678", "patient id: 5678"), accept and use it — do not ask again.
+- Avoid repeating unnecessary clarifying questions. If the user has already given the PID, proceed to use it. If you previously asked for the PID and the user didn't provide it, ask once more succinctly and then offer to help with general guidance without records.
 
 STRICT OUTPUT FORMAT (JSON ONLY):
 Return a single JSON object with the following keys:
@@ -83,7 +82,52 @@ Rules:
 - Do not make up information that is not present in the provided medical reports or conversation history.
 """
 
-# --- JSON extraction helper ---
+# --- Helper utilities ---
+PID_PATTERN = re.compile(r"(?:\bpid\b|\bpatient\s*id\b|\bp\b)\s*[:#\-]?\s*(p?\d+)", re.IGNORECASE)
+DIGIT_PATTERN = re.compile(r"\b(p?\d{3,})\b")  # fallback: any 3+ digit cluster
+
+RECORD_KEYWORDS = [
+    "report", "lab", "result", "results", "previous", "history", "record", "records",
+    "test", "tests", "scan", "imaging", "radiology", "thyroid", "tsh", "t3", "t4",
+    "prescription", "doctor", "referral", "visit", "consultation",
+]
+
+
+def extract_pid_from_text(text: str) -> str | None:
+    """Try to extract PID from a free-form text string.
+    Accepts patterns like: 'pid 5678', 'p5678', 'patient id: 5678', or a bare number if clearly intended as an id.
+    """
+    if not text:
+        return None
+    m = PID_PATTERN.search(text)
+    if m:
+        return m.group(1).lstrip('pP')
+    # fallback: look for explicit mention like 'pid p5678' with different casing
+    # final fallback: any 3+ digit cluster but only if message also contains a record keyword
+    if any(k in text.lower() for k in RECORD_KEYWORDS):
+        m2 = DIGIT_PATTERN.search(text)
+        if m2:
+            return m2.group(1).lstrip('pP')
+    return None
+
+
+def needs_pid_for_query(text: str) -> bool:
+    """Decide whether the user's message requires looking up prior records (thus needs a PID).
+    Simple heuristic: if message contains any keyword from RECORD_KEYWORDS or explicit phrases asking for previous tests/records.
+    """
+    if not text:
+        return False
+    lower = text.lower()
+    # direct phrases that clearly require historical records
+    phrases = ["previous report", "previous lab", "my report", "my records", "past report", "last report", "previous test", "previous results"]
+    if any(p in lower for p in phrases):
+        return True
+    # if message contains any record-related keyword, treat as needing PID
+    if any(k in lower for k in RECORD_KEYWORDS):
+        return True
+    return False
+
+# --- JSON extraction helper (unchanged) ---
 def extract_json_from_llm_response(raw_response: str) -> dict:
     """Safely extracts a JSON object from a string that might contain extra text or markdown."""
     default = {
@@ -135,14 +179,14 @@ def serve_frontend():
     try:
         return app.send_static_file("frontend.html")
     except Exception:
-        return "<h3>frontend2.html not found in static/ — please add your frontend2.html there.</h3>", 404
+        return "<h3>frontend.html not found in static/ — please add your frontend.html there.</h3>", 404
 
 @app.route("/upload_report", methods=["POST"])
 def upload_report():
     """Handles the upload of a new PDF report for a specific patient."""
     if 'report' not in request.files:
         return jsonify({"error": "No file part in the request"}), 400
-    
+
     file = request.files['report']
     patient_id = request.form.get("patient_id")
 
@@ -167,9 +211,11 @@ def chat():
     chat_history = data.get("chat_history") or []
     patient_state = data.get("patient_state") or {}
     patient_id = patient_state.get("patientDetails", {}).get("pid")
-    
+
     # --- Prepare the state for the LLM ---
     state = patient_state.copy()
+    state.setdefault("asked_for_pid", False)
+    state.setdefault("conversationSummary", state.get("conversationSummary", ""))
     state["lastUserMessage"] = ""
     if chat_history:
         # Find the last user message
@@ -178,25 +224,18 @@ def chat():
                 state["lastUserMessage"] = msg["content"]
                 break
 
+    # Try to infer PID from the last message (user might include it inline)
+    inferred_pid = extract_pid_from_text(state.get("lastUserMessage", "") or "")
+    if not patient_id and inferred_pid:
+        logger.info("Inferred PID from user message: %s", inferred_pid)
+        state.setdefault("patientDetails", {})["pid"] = inferred_pid
+        patient_id = inferred_pid
+
     combined_text_parts = []
-    # If a PID is not yet known, prompt the agent to ask for it.
-    if not patient_id:
-        # A simple prompt to get the agent to ask for the PID.
-        user_prompt = "Hello. I need to get the patient's ID to proceed."
-        
-        # Check if the user's last message contains a possible number for the PID
-        last_message = state.get("lastUserMessage", "")
-        # A very basic check to see if the user provided a number
-        if re.search(r'\d+', last_message):
-            inferred_pid = re.search(r'(\d+)', last_message).group(1)
-            state["patientDetails"] = {"pid": inferred_pid}
-            patient_id = inferred_pid
-            # Now that we have a PID, let the agent know to process the reports.
-            user_prompt = f"The user provided a patient ID: {inferred_pid}. Please access their reports and respond."
-        else:
-            # If no PID is found, the agent should ask for it.
-            user_prompt = "The patient has not provided a patient ID. Please ask them to provide it to proceed."
-    
+
+    # Decide whether this query actually needs patient records
+    wants_records = needs_pid_for_query(state.get("lastUserMessage", "") or "")
+
     # If a PID is known, load the patient reports.
     if patient_id:
         patient_folder = REPORTS_ROOT / f"p_{patient_id}"
@@ -215,12 +254,12 @@ def chat():
                         page_text = file_path.read_text(encoding='utf-8', errors='ignore')
                     except Exception:
                         page_text = ""
-                
+
                 if page_text:
                     cleaned = clean_notes_with_bloatectomy(page_text, style="remov")
                     if cleaned:
                         combined_text_parts.append(cleaned)
-    
+
     # Update the conversation summary with the parsed documents
     base_summary = state.get("conversationSummary", "") or ""
     docs_summary = "\n\n".join(combined_text_parts)
@@ -229,12 +268,26 @@ def chat():
     else:
         state["conversationSummary"] = base_summary
 
-    # --- Direct LLM Invocation ---
+    # Build the user prompt for the LLM. We explicitly tell the LLM whether a PID is available and whether the user's message requires records.
+    if patient_id:
+        action_hint = f"Use the patient ID {patient_id} to retrieve and summarize any relevant reports."
+    else:
+        if wants_records and not state.get("asked_for_pid", False):
+            action_hint = "The user is asking for information that requires prior medical records. Ask succinctly for the Patient ID (PID) if needed."
+            # mark that we asked for PID so we don't repeatedly ask
+            state["asked_for_pid"] = True
+        elif wants_records and state.get("asked_for_pid", False):
+            action_hint = "The user previously was asked for a PID but has not supplied one. Ask once more concisely for the PID; otherwise offer to help with general guidance without accessing records."
+        else:
+            action_hint = "No PID provided and the user's request does not need prior records. Provide helpful, general medical guidance and offer to retrieve records if the user later supplies a PID."
+
     user_prompt = f"""
 Current patientDetails: {json.dumps(state.get("patientDetails", {}))}
-Current conversationSummary: {state.get("conversationSummary", "")}
+Current conversationSummary: {state.get("conversationSummary", "")[:4000]}
 Last user message: {state.get("lastUserMessage", "")}
-    
+
+SYSTEM_HINT: {action_hint}
+
 Return ONLY valid JSON with keys: assistant_reply, patientDetails, conversationSummary.
 """
 
@@ -242,7 +295,7 @@ Return ONLY valid JSON with keys: assistant_reply, patientDetails, conversationS
         {"role": "system", "content": PATIENT_ASSISTANT_PROMPT},
         {"role": "user", "content": user_prompt}
     ]
-    
+
     try:
         logger.info("Invoking LLM with prepared state and prompt...")
         llm_response = llm.invoke(messages)
@@ -254,21 +307,26 @@ Return ONLY valid JSON with keys: assistant_reply, patientDetails, conversationS
 
         logger.info(f"Raw LLM response: {raw_response}")
         parsed_result = extract_json_from_llm_response(raw_response)
-        
+
     except Exception as e:
         logger.exception("LLM invocation failed")
         return jsonify({"error": "LLM invocation failed", "detail": str(e)}), 500
 
     updated_state = parsed_result or {}
 
+    # Merge patientDetails back into state (but avoid overwriting asked_for_pid)
+    state.setdefault("patientDetails", {}).update(updated_state.get("patientDetails", {}))
+    state["conversationSummary"] = updated_state.get("conversationSummary", state.get("conversationSummary", ""))
+
     assistant_reply = updated_state.get("assistant_reply")
     if not assistant_reply or not isinstance(assistant_reply, str) or not assistant_reply.strip():
         # Fallback to a polite message if the LLM response is invalid or empty
-        assistant_reply = "I'm here to help — could you tell me more about your symptoms?"
+        assistant_reply = "I'm here to help — could you tell me more about your symptoms or provide a Patient ID (PID) if you want me to fetch past reports?"
 
+    # Return the new assistant reply and the updated state so the frontend can persist it.
     response_payload = {
         "assistant_reply": assistant_reply,
-        "updated_state": updated_state,
+        "updated_state": state,
     }
 
     return jsonify(response_payload)
@@ -281,12 +339,6 @@ def upload_reports():
     Expects multipart/form-data with:
       - patient_id (form field)
       - files (one or multiple files; use the same field name 'files' for each file)
-
-    Example curl:
-    curl -X POST http://localhost:7860/upload_reports \
-      -F "patient_id=12345" \
-      -F "files[]=@/path/to/report1.pdf" \
-      -F "files[]=@/path/to/report2.pdf"
     """
     try:
         # patient id can be in form or args (for convenience)
@@ -365,4 +417,4 @@ def ping():
 
 if __name__ == "__main__":
     port = int(os.getenv("PORT", 7860))
-    app.run(host="0.0.0.0", port=port, debug=True)
+    app.run(host="0.0.0.0", port=port, debug=True)