import streamlit as st import requests import json import re import os import operator import traceback from functools import lru_cache from dotenv import load_dotenv from langchain_groq import ChatGroq from langchain_community.tools.tavily_search import TavilySearchResults from langchain_core.messages import HumanMessage, SystemMessage, AIMessage, ToolMessage from langchain_core.prompts import ChatPromptTemplate from langchain_core.pydantic_v1 import BaseModel, Field from langchain_core.tools import tool from langgraph.prebuilt import ToolExecutor from langgraph.graph import StateGraph, END from typing import Optional, List, Dict, Any, TypedDict, Annotated # --- Environment Variable Loading & Validation --- load_dotenv() # Load .env file if present (for local development) UMLS_API_KEY = os.environ.get("UMLS_API_KEY") GROQ_API_KEY = os.environ.get("GROQ_API_KEY") TAVILY_API_KEY = os.environ.get("TAVILY_API_KEY") # Stop execution if essential keys are missing (crucial for HF Spaces) missing_keys = [] if not UMLS_API_KEY: missing_keys.append("UMLS_API_KEY") if not GROQ_API_KEY: missing_keys.append("GROQ_API_KEY") if not TAVILY_API_KEY: missing_keys.append("TAVILY_API_KEY") if missing_keys: st.error(f"Missing required API Key(s): {', '.join(missing_keys)}. Please set them in Hugging Face Space Secrets or your environment variables.") st.stop() # --- Configuration & Constants --- class ClinicalAppSettings: APP_TITLE = "SynapseAI: Interactive Clinical Decision Support (UMLS/FDA Integrated)" PAGE_LAYOUT = "wide" MODEL_NAME = "llama3-70b-8192" # Groq Llama3 70b TEMPERATURE = 0.1 MAX_SEARCH_RESULTS = 3 class ClinicalPrompts: # System prompt remains the same as the previous version, emphasizing structured output, # safety checks, guideline search, and conversational flow. SYSTEM_PROMPT = """ You are SynapseAI, an expert AI clinical assistant engaged in an interactive consultation. Your goal is to support healthcare professionals by analyzing patient data, providing differential diagnoses, suggesting evidence-based management plans, and identifying risks according to current standards of care. **Core Directives for this Conversation:** 1. **Analyze Sequentially:** Process information turn-by-turn. Base your responses on the *entire* conversation history. 2. **Seek Clarity:** If the provided information is insufficient or ambiguous for a safe assessment, CLEARLY STATE what specific additional information or clarification is needed. Do NOT guess or make unsafe assumptions. 3. **Structured Assessment (When Ready):** When you have sufficient information and have performed necessary checks (like interactions, guideline searches), provide a comprehensive assessment using the following JSON structure. Output this JSON structure as the primary content of your response when you are providing the full analysis. Do NOT output incomplete JSON. If you need to ask a question or perform a tool call first, do that instead of outputting this structure. ```json { "assessment": "Concise summary of the patient's presentation and key findings based on the conversation.", "differential_diagnosis": [ {"diagnosis": "Primary Diagnosis", "likelihood": "High/Medium/Low", "rationale": "Supporting evidence from conversation..."}, {"diagnosis": "Alternative Diagnosis 1", "likelihood": "Medium/Low", "rationale": "Supporting/Refuting evidence..."}, {"diagnosis": "Alternative Diagnosis 2", "likelihood": "Low", "rationale": "Why it's less likely but considered..."} ], "risk_assessment": { "identified_red_flags": ["List any triggered red flags based on input and analysis"], "immediate_concerns": ["Specific urgent issues requiring attention (e.g., sepsis risk, ACS rule-out)"], "potential_complications": ["Possible future issues based on presentation"] }, "recommended_plan": { "investigations": ["List specific lab tests or imaging required. Use 'order_lab_test' tool."], "therapeutics": ["Suggest specific treatments or prescriptions. Use 'prescribe_medication' tool. MUST check interactions first using 'check_drug_interactions'."], "consultations": ["Recommend specialist consultations if needed."], "patient_education": ["Key points for patient communication."] }, "rationale_summary": "Justification for assessment/plan. **Crucially, if relevant (e.g., ACS, sepsis, common infections), use 'tavily_search_results' to find and cite current clinical practice guidelines (e.g., 'latest ACC/AHA chest pain guidelines 202X', 'Surviving Sepsis Campaign guidelines') supporting your recommendations.** Include summary of guideline findings here.", "interaction_check_summary": "Summary of findings from 'check_drug_interactions' if performed." } ``` 4. **Safety First - Interactions:** BEFORE suggesting a new prescription via `prescribe_medication`, you MUST FIRST use `check_drug_interactions` in a preceding or concurrent tool call. Report the findings from the interaction check. If significant interactions exist, modify the plan or state the contraindication clearly. 5. **Safety First - Red Flags:** Use the `flag_risk` tool IMMEDIATELY if critical red flags requiring urgent action are identified at any point in the conversation. 6. **Tool Use:** Employ tools (`order_lab_test`, `prescribe_medication`, `check_drug_interactions`, `flag_risk`, `tavily_search_results`) logically within the conversational flow. Wait for tool results before proceeding if the result is needed for the next step (e.g., wait for interaction check before confirming prescription in the structured JSON). 7. **Evidence & Guidelines:** Actively use `tavily_search_results` not just for general knowledge, but specifically to query for and incorporate **current clinical practice guidelines** relevant to the patient's presentation (e.g., chest pain, shortness of breath, suspected infection). Summarize findings in the `rationale_summary` when providing the structured output. 8. **Conciseness & Flow:** Be medically accurate and concise. Use standard terminology. Respond naturally in conversation (asking questions, acknowledging info) until ready for the full structured JSON output. """ # --- UMLS/RxNorm & OpenFDA API Helper Functions --- UMLS_AUTH_ENDPOINT = "https://utslogin.nlm.nih.gov/cas/v1/api-key" # May not be needed if using apiKey directly RXNORM_API_BASE = "https://rxnav.nlm.nih.gov/REST" OPENFDA_API_BASE = "https://api.fda.gov/drug/label.json" @lru_cache(maxsize=256) # Cache RxCUI lookups def get_rxcui(drug_name: str) -> Optional[str]: """Uses RxNorm API to find the RxCUI for a given drug name.""" if not drug_name or not isinstance(drug_name, str): return None drug_name = drug_name.strip() if not drug_name: return None print(f"RxNorm Lookup for: '{drug_name}'") try: params = {"name": drug_name, "search": 1} # Search for concepts related to the name response = requests.get(f"{RXNORM_API_BASE}/rxcui.json", params=params, timeout=10) response.raise_for_status() data = response.json() # Extract RxCUI - prioritize exact matches or common types if data and "idGroup" in data and "rxnormId" in data["idGroup"]: # Select the first one, assuming it's the most relevant by default. # More sophisticated logic could check TTYs (Term Types) if needed. rxcui = data["idGroup"]["rxnormId"][0] print(f" Found RxCUI: {rxcui} for '{drug_name}'") return rxcui else: # Fallback: Search /drugs endpoint if direct rxcui lookup fails params = {"name": drug_name} response = requests.get(f"{RXNORM_API_BASE}/drugs.json", params=params, timeout=10) response.raise_for_status() data = response.json() if data and "drugGroup" in data and "conceptGroup" in data["drugGroup"]: for group in data["drugGroup"]["conceptGroup"]: # Prioritize Semantic Types like Brand/Clinical Drug/Ingredient if group.get("tty") in ["SBD", "SCD", "GPCK", "BPCK", "IN", "MIN", "PIN"]: if "conceptProperties" in group and group["conceptProperties"]: rxcui = group["conceptProperties"][0].get("rxcui") if rxcui: print(f" Found RxCUI (via /drugs): {rxcui} for '{drug_name}'") return rxcui print(f" RxCUI not found for '{drug_name}'.") return None except requests.exceptions.RequestException as e: print(f" Error fetching RxCUI for '{drug_name}': {e}") return None except json.JSONDecodeError as e: print(f" Error decoding RxNorm JSON response for '{drug_name}': {e}") return None except Exception as e: # Catch any other unexpected error print(f" Unexpected error in get_rxcui for '{drug_name}': {e}") return None @lru_cache(maxsize=128) # Cache OpenFDA lookups def get_openfda_label(rxcui: Optional[str] = None, drug_name: Optional[str] = None) -> Optional[dict]: """Fetches drug label information from OpenFDA using RxCUI or drug name.""" if not rxcui and not drug_name: return None print(f"OpenFDA Label Lookup for: RXCUI={rxcui}, Name={drug_name}") search_terms = [] # Prioritize RxCUI lookup using multiple potential fields if rxcui: search_terms.append(f'spl_rxnorm_code:"{rxcui}"') search_terms.append(f'openfda.rxcui:"{rxcui}"') # Add name search as fallback or supplement if drug_name: search_terms.append(f'(openfda.brand_name:"{drug_name.lower()}" OR openfda.generic_name:"{drug_name.lower()}")') search_query = " OR ".join(search_terms) params = {"search": search_query, "limit": 1} # Get only the most relevant label try: response = requests.get(OPENFDA_API_BASE, params=params, timeout=15) response.raise_for_status() data = response.json() if data and "results" in data and data["results"]: print(f" Found OpenFDA label for query: {search_query}") return data["results"][0] # Return the first label found print(f" No OpenFDA label found for query: {search_query}") return None except requests.exceptions.RequestException as e: print(f" Error fetching OpenFDA label: {e}") return None except json.JSONDecodeError as e: print(f" Error decoding OpenFDA JSON response: {e}") return None except Exception as e: print(f" Unexpected error in get_openfda_label: {e}") return None def search_text_list(text_list: Optional[List[str]], search_terms: List[str]) -> List[str]: """ Case-insensitive search for any search_term within a list of text strings. Returns snippets. """ found_snippets = [] if not text_list or not search_terms: return found_snippets # Ensure search terms are lowercased strings search_terms_lower = [str(term).lower() for term in search_terms if term] for text_item in text_list: if not isinstance(text_item, str): continue # Skip non-string items text_item_lower = text_item.lower() for term in search_terms_lower: if term in text_item_lower: # Find the start index of the term start_index = text_item_lower.find(term) # Define snippet boundaries (e.g., 50 chars before, 100 after) snippet_start = max(0, start_index - 50) snippet_end = min(len(text_item), start_index + len(term) + 100) snippet = text_item[snippet_start:snippet_end] # Add indication of where the match is snippet = snippet.replace(term, f"**{term}**", 1) # Highlight first match found_snippets.append(f"...{snippet}...") break # Move to the next text item once a match is found return found_snippets # --- Other Helper Functions --- def parse_bp(bp_string: str) -> Optional[tuple[int, int]]: """Parses BP string like '120/80' into (systolic, diastolic) integers.""" if not isinstance(bp_string, str): return None match = re.match(r"(\d{1,3})\s*/\s*(\d{1,3})", bp_string.strip()) if match: return int(match.group(1)), int(match.group(2)) return None def check_red_flags(patient_data: dict) -> List[str]: """Checks patient data against predefined red flags.""" # (Keep the implementation from the previous full code listing) flags = [] if not patient_data: return flags symptoms = patient_data.get("hpi", {}).get("symptoms", []) vitals = patient_data.get("vitals", {}) history = patient_data.get("pmh", {}).get("conditions", "") symptoms_lower = [str(s).lower() for s in symptoms if isinstance(s, str)] if "chest pain" in symptoms_lower: flags.append("Red Flag: Chest Pain reported.") if "shortness of breath" in symptoms_lower: flags.append("Red Flag: Shortness of Breath reported.") if "severe headache" in symptoms_lower: flags.append("Red Flag: Severe Headache reported.") if "sudden vision loss" in symptoms_lower: flags.append("Red Flag: Sudden Vision Loss reported.") if "weakness on one side" in symptoms_lower: flags.append("Red Flag: Unilateral Weakness reported (potential stroke).") if "hemoptysis" in symptoms_lower: flags.append("Red Flag: Hemoptysis (coughing up blood).") if "syncope" in symptoms_lower: flags.append("Red Flag: Syncope (fainting).") if vitals: temp = vitals.get("temp_c"); hr = vitals.get("hr_bpm"); rr = vitals.get("rr_rpm") spo2 = vitals.get("spo2_percent"); bp_str = vitals.get("bp_mmhg") if temp is not None and temp >= 38.5: flags.append(f"Red Flag: Fever ({temp}°C).") if hr is not None and hr >= 120: flags.append(f"Red Flag: Tachycardia ({hr} bpm).") if hr is not None and hr <= 50: flags.append(f"Red Flag: Bradycardia ({hr} bpm).") if rr is not None and rr >= 24: flags.append(f"Red Flag: Tachypnea ({rr} rpm).") if spo2 is not None and spo2 <= 92: flags.append(f"Red Flag: Hypoxia ({spo2}%).") if bp_str: bp = parse_bp(bp_str) if bp: if bp[0] >= 180 or bp[1] >= 110: flags.append(f"Red Flag: Hypertensive Urgency/Emergency (BP: {bp_str} mmHg).") if bp[0] <= 90 or bp[1] <= 60: flags.append(f"Red Flag: Hypotension (BP: {bp_str} mmHg).") if history and isinstance(history, str): history_lower = history.lower() if "history of mi" in history_lower and "chest pain" in symptoms_lower: flags.append("Red Flag: History of MI with current Chest Pain.") if "history of dvt/pe" in history_lower and "shortness of breath" in symptoms_lower: flags.append("Red Flag: History of DVT/PE with current Shortness of Breath.") return list(set(flags)) # Unique flags def format_patient_data_for_prompt(data: dict) -> str: """Formats the patient dictionary into a readable string for the LLM.""" # (Keep the implementation from the previous full code listing) if not data: return "No patient data provided." prompt_str = "" for key, value in data.items(): section_title = key.replace('_', ' ').title() if isinstance(value, dict) and value: has_content = any(sub_value for sub_value in value.values()) if has_content: prompt_str += f"**{section_title}:**\n" for sub_key, sub_value in value.items(): if sub_value: prompt_str += f" - {sub_key.replace('_', ' ').title()}: {sub_value}\n" elif isinstance(value, list) and value: prompt_str += f"**{section_title}:** {', '.join(map(str, value))}\n" elif value and not isinstance(value, dict): prompt_str += f"**{section_title}:** {value}\n" return prompt_str.strip() # --- Tool Definitions --- # Pydantic models for tool inputs class LabOrderInput(BaseModel): test_name: str = Field(..., description="Specific name of the lab test or panel (e.g., 'CBC', 'BMP', 'Troponin I', 'Urinalysis', 'D-dimer').") reason: str = Field(..., description="Clinical justification for ordering the test (e.g., 'Rule out infection', 'Assess renal function', 'Evaluate for ACS', 'Assess for PE').") priority: str = Field("Routine", description="Priority of the test (e.g., 'STAT', 'Routine').") class PrescriptionInput(BaseModel): medication_name: str = Field(..., description="Name of the medication.") dosage: str = Field(..., description="Dosage amount and unit (e.g., '500 mg', '10 mg', '81 mg').") route: str = Field(..., description="Route of administration (e.g., 'PO', 'IV', 'IM', 'Topical', 'SL').") frequency: str = Field(..., description="How often the medication should be taken (e.g., 'BID', 'QDaily', 'Q4-6H PRN', 'once').") duration: str = Field("As directed", description="Duration of treatment (e.g., '7 days', '1 month', 'Ongoing', 'Until follow-up').") reason: str = Field(..., description="Clinical indication for the prescription.") # Updated InteractionCheckInput - Note: current_medications/allergies are Optional here # because they are populated by the tool_node from state *before* execution. class InteractionCheckInput(BaseModel): potential_prescription: str = Field(..., description="The name of the NEW medication being considered for prescribing.") current_medications: Optional[List[str]] = Field(None, description="List of patient's current medication names (populated from state).") allergies: Optional[List[str]] = Field(None, description="List of patient's known allergies (populated from state).") class FlagRiskInput(BaseModel): risk_description: str = Field(..., description="Specific critical risk identified (e.g., 'Suspected Sepsis', 'Acute Coronary Syndrome', 'Stroke Alert').") urgency: str = Field("High", description="Urgency level (e.g., 'Critical', 'High', 'Moderate').") # Tool functions @tool("order_lab_test", args_schema=LabOrderInput) def order_lab_test(test_name: str, reason: str, priority: str = "Routine") -> str: """Orders a specific lab test with clinical justification and priority.""" print(f"Executing order_lab_test: {test_name}, Reason: {reason}, Priority: {priority}") return json.dumps({"status": "success", "message": f"Lab Ordered: {test_name} ({priority})", "details": f"Reason: {reason}"}) @tool("prescribe_medication", args_schema=PrescriptionInput) def prescribe_medication(medication_name: str, dosage: str, route: str, frequency: str, duration: str, reason: str) -> str: """Prescribes a medication with detailed instructions and clinical indication. IMPORTANT: Requires prior interaction check.""" print(f"Executing prescribe_medication: {medication_name} {dosage}...") # Safety check happens in tool_node *before* this is called. return json.dumps({"status": "success", "message": f"Prescription Prepared: {medication_name} {dosage} {route} {frequency}", "details": f"Duration: {duration}. Reason: {reason}"}) # --- NEW Interaction Check Tool using UMLS/RxNorm & OpenFDA --- @tool("check_drug_interactions", args_schema=InteractionCheckInput) def check_drug_interactions(potential_prescription: str, current_medications: Optional[List[str]] = None, allergies: Optional[List[str]] = None) -> str: """ Checks for potential drug-drug and drug-allergy interactions using RxNorm API for normalization and OpenFDA drug labels for interaction/warning text. REQUIRES UMLS_API_KEY environment variable. """ print(f"\n--- Executing REAL check_drug_interactions ---") print(f"Checking potential prescription: '{potential_prescription}'") warnings = [] potential_med_lower = potential_prescription.lower().strip() # Use provided lists or default to empty current_meds_list = current_medications or [] allergies_list = allergies or [] # Clean and lowercase current med names (basic extraction: first word) current_med_names_lower = [] for med in current_meds_list: match = re.match(r"^\s*([a-zA-Z\-]+)", str(med)) if match: current_med_names_lower.append(match.group(1).lower()) # Clean and lowercase allergies allergies_lower = [str(a).lower().strip() for a in allergies_list if a] print(f" Against Current Meds (names): {current_med_names_lower}") print(f" Against Allergies: {allergies_lower}") # --- Step 1: Normalize potential prescription --- print(f" Step 1: Normalizing '{potential_prescription}'...") potential_rxcui = get_rxcui(potential_prescription) potential_label = get_openfda_label(rxcui=potential_rxcui, drug_name=potential_prescription) if not potential_rxcui and not potential_label: print(f" Warning: Could not find RxCUI or OpenFDA label for '{potential_prescription}'. Interaction check will be limited.") warnings.append(f"INFO: Could not reliably identify '{potential_prescription}' in standard terminologies/databases. Checks may be incomplete.") # --- Step 2: Allergy Check --- print(" Step 2: Performing Allergy Check...") # Direct name match against patient's allergy list for allergy in allergies_lower: if allergy == potential_med_lower: warnings.append(f"CRITICAL ALLERGY (Name Match): Patient allergic to '{allergy}'. Potential prescription is '{potential_prescription}'.") # Basic cross-reactivity check (can be expanded) elif allergy in ["penicillin", "pcns"] and potential_med_lower in ["amoxicillin", "ampicillin", "augmentin", "piperacillin"]: warnings.append(f"POTENTIAL CROSS-ALLERGY: Patient allergic to Penicillin. High risk with '{potential_prescription}'.") elif allergy == "sulfa" and potential_med_lower in ["sulfamethoxazole", "bactrim", "sulfasalazine"]: warnings.append(f"POTENTIAL CROSS-ALLERGY: Patient allergic to Sulfa. High risk with '{potential_prescription}'.") elif allergy in ["nsaids", "aspirin"] and potential_med_lower in ["ibuprofen", "naproxen", "ketorolac", "diclofenac"]: warnings.append(f"POTENTIAL CROSS-ALLERGY: Patient allergic to NSAIDs/Aspirin. Risk with '{potential_prescription}'.") # Check OpenFDA Label for Contraindications/Warnings related to ALLERGIES if potential_label: contraindications = potential_label.get("contraindications") warnings_section = potential_label.get("warnings_and_cautions") or potential_label.get("warnings") if contraindications: allergy_mentions_ci = search_text_list(contraindications, allergies_lower) if allergy_mentions_ci: warnings.append(f"ALLERGY RISK (Contraindication Found): Label for '{potential_prescription}' mentions contraindication potentially related to patient allergies: {'; '.join(allergy_mentions_ci)}") if warnings_section: allergy_mentions_warn = search_text_list(warnings_section, allergies_lower) if allergy_mentions_warn: warnings.append(f"ALLERGY RISK (Warning Found): Label for '{potential_prescription}' mentions warnings potentially related to patient allergies: {'; '.join(allergy_mentions_warn)}") # --- Step 3: Drug-Drug Interaction Check --- print(" Step 3: Performing Drug-Drug Interaction Check...") if potential_rxcui or potential_label: # Proceed only if we have info on the potential drug for current_med_name in current_med_names_lower: if not current_med_name or current_med_name == potential_med_lower: continue # Skip empty or self-interaction print(f" Checking interaction between '{potential_prescription}' and '{current_med_name}'...") current_rxcui = get_rxcui(current_med_name) current_label = get_openfda_label(rxcui=current_rxcui, drug_name=current_med_name) # Terms to search for in interaction text search_terms_for_current = [current_med_name] if current_rxcui: search_terms_for_current.append(current_rxcui) # Add RxCUI if found search_terms_for_potential = [potential_med_lower] if potential_rxcui: search_terms_for_potential.append(potential_rxcui) # Add RxCUI if found interaction_found_flag = False # Check Potential Drug's Label ('drug_interactions' section) for mentions of Current Drug if potential_label and potential_label.get("drug_interactions"): interaction_mentions = search_text_list(potential_label.get("drug_interactions"), search_terms_for_current) if interaction_mentions: warnings.append(f"Potential Interaction ({potential_prescription.capitalize()} Label): Mentions '{current_med_name.capitalize()}'. Snippets: {'; '.join(interaction_mentions)}") interaction_found_flag = True # Check Current Drug's Label ('drug_interactions' section) for mentions of Potential Drug if current_label and current_label.get("drug_interactions") and not interaction_found_flag: # Avoid duplicate warnings if already found interaction_mentions = search_text_list(current_label.get("drug_interactions"), search_terms_for_potential) if interaction_mentions: warnings.append(f"Potential Interaction ({current_med_name.capitalize()} Label): Mentions '{potential_prescription.capitalize()}'. Snippets: {'; '.join(interaction_mentions)}") else: # Case where potential drug wasn't identified warnings.append(f"INFO: Drug-drug interaction check skipped for '{potential_prescription}' as it could not be identified via RxNorm/OpenFDA.") # --- Step 4: Format Output --- final_warnings = list(set(warnings)) # Remove duplicates status = "warning" if any("CRITICAL" in w or "Interaction" in w or "RISK" in w for w in final_warnings) else "clear" if not final_warnings: status = "clear" # Ensure clear if no warnings remain message = f"Interaction/Allergy check for '{potential_prescription}': {len(final_warnings)} potential issue(s) identified using RxNorm/OpenFDA." if final_warnings else f"No major interactions or allergy issues identified for '{potential_prescription}' based on RxNorm/OpenFDA lookup." print(f"--- Interaction Check Complete for '{potential_prescription}' ---") return json.dumps({"status": status, "message": message, "warnings": final_warnings}) # --- End of NEW Interaction Check Tool --- @tool("flag_risk", args_schema=FlagRiskInput) def flag_risk(risk_description: str, urgency: str) -> str: """Flags a critical risk identified during analysis for immediate attention.""" print(f"Executing flag_risk: {risk_description}, Urgency: {urgency}") st.error(f"🚨 **{urgency.upper()} RISK FLAGGED by AI:** {risk_description}", icon="🚨") return json.dumps({"status": "flagged", "message": f"Risk '{risk_description}' flagged with {urgency} urgency."}) # Initialize Search Tool search_tool = TavilySearchResults( max_results=ClinicalAppSettings.MAX_SEARCH_RESULTS, name="tavily_search_results" ) # --- LangGraph Setup --- # Define the state structure class AgentState(TypedDict): messages: Annotated[list[Any], operator.add] patient_data: Optional[dict] # Define Tools and Tool Executor tools = [ order_lab_test, prescribe_medication, check_drug_interactions, # Using the new implementation flag_risk, search_tool ] tool_executor = ToolExecutor(tools) # Define the Agent Model model = ChatGroq( temperature=ClinicalAppSettings.TEMPERATURE, model=ClinicalAppSettings.MODEL_NAME, ) model_with_tools = model.bind_tools(tools) # --- Graph Nodes (agent_node, tool_node remain mostly the same structurally) --- # 1. Agent Node: Calls the LLM (No change needed from previous version) def agent_node(state: AgentState): """Invokes the LLM to decide the next action or response.""" print("\n---AGENT NODE---") current_messages = state['messages'] if not current_messages or not isinstance(current_messages[0], SystemMessage): print("Prepending System Prompt.") current_messages = [SystemMessage(content=ClinicalPrompts.SYSTEM_PROMPT)] + current_messages print(f"Invoking LLM with {len(current_messages)} messages.") try: response = model_with_tools.invoke(current_messages) print(f"Agent Raw Response Type: {type(response)}") if hasattr(response, 'tool_calls') and response.tool_calls: print(f"Agent Response Tool Calls: {response.tool_calls}") else: print("Agent Response: No tool calls.") except Exception as e: print(f"ERROR in agent_node during LLM invocation: {type(e).__name__} - {e}") traceback.print_exc() error_message = AIMessage(content=f"Sorry, an internal error occurred while processing the request: {type(e).__name__}") return {"messages": [error_message]} return {"messages": [response]} # 2. Tool Node: Executes tools (Mostly the same, ensures context injection) def tool_node(state: AgentState): """Executes tools called by the LLM and returns results.""" print("\n---TOOL NODE---") tool_messages = [] last_message = state['messages'][-1] if not isinstance(last_message, AIMessage) or not getattr(last_message, 'tool_calls', None): print("Warning: Tool node called unexpectedly without tool calls.") return {"messages": []} tool_calls = last_message.tool_calls print(f"Tool calls received: {json.dumps(tool_calls, indent=2)}") # Safety Check Logic (No change needed from previous version) prescriptions_requested = {} interaction_checks_requested = {} for call in tool_calls: tool_name = call.get('name'); tool_args = call.get('args', {}) if tool_name == 'prescribe_medication': med_name = tool_args.get('medication_name', '').lower(); if med_name: prescriptions_requested[med_name] = call elif tool_name == 'check_drug_interactions': potential_med = tool_args.get('potential_prescription', '').lower() if potential_med: interaction_checks_requested[potential_med] = call valid_tool_calls_for_execution = [] blocked_ids = set() for med_name, prescribe_call in prescriptions_requested.items(): if med_name not in interaction_checks_requested: st.error(f"**Safety Violation:** AI attempted to prescribe '{med_name}' without requesting `check_drug_interactions` in the *same turn*. Prescription blocked.") error_msg = ToolMessage(content=json.dumps({"status": "error", "message": f"Interaction check for '{med_name}' must be requested *before or alongside* the prescription call."}), tool_call_id=prescribe_call['id'], name=prescribe_call['name']) tool_messages.append(error_msg) blocked_ids.add(prescribe_call['id']) valid_tool_calls_for_execution = [call for call in tool_calls if call['id'] not in blocked_ids] # Augment interaction checks with patient data (Crucial part - no change needed here) patient_data = state.get("patient_data", {}) patient_meds_full = patient_data.get("medications", {}).get("current", []) # Pass full med list if needed by tool patient_allergies = patient_data.get("allergies", []) for call in valid_tool_calls_for_execution: if call['name'] == 'check_drug_interactions': if 'args' not in call: call['args'] = {} # Pass the necessary context from patient_data to the tool arguments # The tool function expects 'current_medications' (list of names) and 'allergies' call['args']['current_medications'] = patient_meds_full # Pass the full strings call['args']['allergies'] = patient_allergies print(f"Augmented interaction check args for call ID {call['id']}") # Removed args content for brevity # Execute valid tool calls (No change needed from previous version) if valid_tool_calls_for_execution: print(f"Attempting to execute {len(valid_tool_calls_for_execution)} tools: {[c['name'] for c in valid_tool_calls_for_execution]}") try: responses = tool_executor.batch(valid_tool_calls_for_execution, return_exceptions=True) for call, resp in zip(valid_tool_calls_for_execution, responses): tool_call_id = call['id']; tool_name = call['name'] if isinstance(resp, Exception): error_type = type(resp).__name__; error_str = str(resp) print(f"ERROR executing tool '{tool_name}' (ID: {tool_call_id}): {error_type} - {error_str}") traceback.print_exc() st.error(f"Error executing action '{tool_name}': {error_type}") error_content = json.dumps({"status": "error", "message": f"Failed to execute '{tool_name}': {error_type} - {error_str}"}) tool_messages.append(ToolMessage(content=error_content, tool_call_id=tool_call_id, name=tool_name)) if isinstance(resp, AttributeError) and "'dict' object has no attribute 'tool'" in error_str: print("\n *** DETECTED SPECIFIC ATTRIBUTE ERROR ('dict' object has no attribute 'tool') *** \n") else: print(f"Tool '{tool_name}' (ID: {tool_call_id}) executed successfully.") content_str = str(resp) tool_messages.append(ToolMessage(content=content_str, tool_call_id=tool_call_id, name=tool_name)) except Exception as e: print(f"CRITICAL UNEXPECTED ERROR within tool_node logic: {type(e).__name__} - {e}") traceback.print_exc(); st.error(f"Critical internal error processing actions: {e}") error_content = json.dumps({"status": "error", "message": f"Internal error processing tools: {e}"}) processed_ids = {msg.tool_call_id for msg in tool_messages} for call in valid_tool_calls_for_execution: if call['id'] not in processed_ids: tool_messages.append(ToolMessage(content=error_content, tool_call_id=call['id'], name=call['name'])) print(f"Returning {len(tool_messages)} tool messages.") return {"messages": tool_messages} # --- Graph Edges (Routing Logic) --- (No change needed) def should_continue(state: AgentState) -> str: """Determines whether to call tools, end the conversation turn, or handle errors.""" print("\n---ROUTING DECISION---") last_message = state['messages'][-1] if state['messages'] else None if not isinstance(last_message, AIMessage): return "end_conversation_turn" if "Sorry, an internal error occurred" in last_message.content: return "end_conversation_turn" if getattr(last_message, 'tool_calls', None): return "continue_tools" else: return "end_conversation_turn" # --- Graph Definition & Compilation --- (No change needed) workflow = StateGraph(AgentState) workflow.add_node("agent", agent_node) workflow.add_node("tools", tool_node) workflow.set_entry_point("agent") workflow.add_conditional_edges("agent", should_continue, {"continue_tools": "tools", "end_conversation_turn": END}) workflow.add_edge("tools", "agent") app = workflow.compile() print("LangGraph compiled successfully.") # --- Streamlit UI --- def main(): st.set_page_config(page_title=ClinicalAppSettings.APP_TITLE, layout=ClinicalAppSettings.PAGE_LAYOUT) st.title(f"🩺 {ClinicalAppSettings.APP_TITLE}") st.caption(f"Interactive Assistant | LangGraph/Groq/Tavily/UMLS/OpenFDA | Model: {ClinicalAppSettings.MODEL_NAME}") # Initialize session state (No change needed) if "messages" not in st.session_state: st.session_state.messages = [] if "patient_data" not in st.session_state: st.session_state.patient_data = None if "graph_app" not in st.session_state: st.session_state.graph_app = app # --- Patient Data Input Sidebar --- (Adjusted allergy/med extraction slightly) with st.sidebar: st.header("📄 Patient Intake Form") # Demographics, HPI, History, Social/Family, Vitals/Exam sections remain the same input fields # ... (Copy input fields from previous full code version) ... st.subheader("Demographics") age = st.number_input("Age", min_value=0, max_value=120, value=55, key="age_input") sex = st.selectbox("Biological Sex", ["Male", "Female", "Other/Prefer not to say"], key="sex_input") st.subheader("History of Present Illness (HPI)") chief_complaint = st.text_input("Chief Complaint", "Chest pain", key="cc_input") hpi_details = st.text_area("Detailed HPI", "55 y/o male presents with substernal chest pain started 2 hours ago...", key="hpi_input", height=150) symptoms = st.multiselect("Associated Symptoms", ["Nausea", "Diaphoresis", "Shortness of Breath", "Dizziness", "Palpitations", "Fever", "Cough", "Severe Headache", "Syncope", "Hemoptysis"], default=["Nausea", "Diaphoresis"], key="sym_input") st.subheader("Past History") pmh = st.text_area("Past Medical History (PMH)", "Hypertension (HTN), Hyperlipidemia (HLD), Type 2 Diabetes Mellitus (DM2), History of MI", key="pmh_input") psh = st.text_area("Past Surgical History (PSH)", "Appendectomy (2005)", key="psh_input") st.subheader("Medications & Allergies") current_meds_str = st.text_area("Current Medications (name, dose, freq)", "Lisinopril 10mg daily\nMetformin 1000mg BID\nAtorvastatin 40mg daily\nAspirin 81mg daily", key="meds_input") allergies_str = st.text_area("Allergies (comma separated, specify reaction if known)", "Penicillin (rash), Sulfa (hives)", key="allergy_input") st.subheader("Social/Family History") social_history = st.text_area("Social History (SH)", "Smoker (1 ppd x 30 years), occasional alcohol.", key="sh_input") family_history = st.text_area("Family History (FHx)", "Father had MI at age 60. Mother has HTN.", key="fhx_input") st.subheader("Vitals & Exam Findings") col1, col2 = st.columns(2) with col1: temp_c = st.number_input("Temp (°C)", 35.0, 42.0, 36.8, format="%.1f", key="temp_input") hr_bpm = st.number_input("HR (bpm)", 30, 250, 95, key="hr_input") rr_rpm = st.number_input("RR (rpm)", 5, 50, 18, key="rr_input") with col2: bp_mmhg = st.text_input("BP (SYS/DIA)", "155/90", key="bp_input") spo2_percent = st.number_input("SpO2 (%)", 70, 100, 96, key="spo2_input") pain_scale = st.slider("Pain (0-10)", 0, 10, 8, key="pain_input") exam_notes = st.text_area("Brief Physical Exam Notes", "Awake, alert, oriented x3. Mild distress. Lungs clear bilaterally...", key="exam_input", height=100) # Compile Patient Data Dictionary (Refined Extraction for Tool Use) if st.button("Start/Update Consultation", key="start_button"): # Store full medication strings for display/context current_meds_list = [med.strip() for med in current_meds_str.split('\n') if med.strip()] # Extract just the names (simplified) for the interaction check tool's state population current_med_names_only = [] for med in current_meds_list: match = re.match(r"^\s*([a-zA-Z\-]+)", med) if match: current_med_names_only.append(match.group(1).lower()) # Extract allergy names (simplified, before parenthesis) allergies_list = [] for a in allergies_str.split(','): cleaned_allergy = a.strip() if cleaned_allergy: match = re.match(r"^\s*([a-zA-Z\-\s/]+)(?:\s*\(.*\))?", cleaned_allergy) name_part = match.group(1).strip().lower() if match else cleaned_allergy.lower() allergies_list.append(name_part) st.session_state.patient_data = { "demographics": {"age": age, "sex": sex}, "hpi": {"chief_complaint": chief_complaint, "details": hpi_details, "symptoms": symptoms}, "pmh": {"conditions": pmh}, "psh": {"procedures": psh}, # Store both full list and names_only list "medications": {"current": current_meds_list, "names_only": current_med_names_only}, "allergies": allergies_list, # Store cleaned list "social_history": {"details": social_history}, "family_history": {"details": family_history}, "vitals": { "temp_c": temp_c, "hr_bpm": hr_bpm, "bp_mmhg": bp_mmhg, "rr_rpm": rr_rpm, "spo2_percent": spo2_percent, "pain_scale": pain_scale}, "exam_findings": {"notes": exam_notes} } # Initial Red Flag Check red_flags = check_red_flags(st.session_state.patient_data) st.sidebar.markdown("---") if red_flags: st.sidebar.warning("**Initial Red Flags Detected:**") for flag in red_flags: st.sidebar.warning(f"- {flag.replace('Red Flag: ','')}") else: st.sidebar.success("No immediate red flags detected.") # Prepare initial message & reset history initial_prompt = "Initiate consultation for the patient described in the intake form. Review the data and begin analysis." st.session_state.messages = [HumanMessage(content=initial_prompt)] st.success("Patient data loaded/updated. Ready for analysis.") # --- Main Chat Interface Area --- (No change needed in display logic) st.header("💬 Clinical Consultation") # Display chat messages from history # (Copy the message display loop from the previous full code version) for msg_index, msg in enumerate(st.session_state.messages): unique_key = f"msg_{msg_index}" if isinstance(msg, HumanMessage): with st.chat_message("user", key=f"{unique_key}_user"): st.markdown(msg.content) elif isinstance(msg, AIMessage): with st.chat_message("assistant", key=f"{unique_key}_ai"): ai_content = msg.content; structured_output = None try: json_match = re.search(r"```json\s*(\{.*?\})\s*```", ai_content, re.DOTALL | re.IGNORECASE) if json_match: json_str = json_match.group(1); prefix = ai_content[:json_match.start()].strip(); suffix = ai_content[json_match.end():].strip() if prefix: st.markdown(prefix) structured_output = json.loads(json_str) if suffix: st.markdown(suffix) elif ai_content.strip().startswith("{") and ai_content.strip().endswith("}"): structured_output = json.loads(ai_content); ai_content = "" else: st.markdown(ai_content) except Exception as e: st.markdown(ai_content); print(f"Error parsing/displaying AI JSON: {e}") if structured_output and isinstance(structured_output, dict): # (Copy the structured JSON display logic from previous full code) st.divider(); st.subheader("📊 AI Analysis & Recommendations") cols = st.columns(2) with cols[0]: st.markdown(f"**Assessment:**"); st.markdown(f"> {structured_output.get('assessment', 'N/A')}") st.markdown(f"**Differential Diagnosis:**") ddx = structured_output.get('differential_diagnosis', []); if ddx: for item in ddx: likelihood = item.get('likelihood', '?').capitalize(); icon = "🥇" if likelihood=="High" else ("🥈" if likelihood=="Medium" else "🥉") with st.expander(f"{icon} {item.get('diagnosis', 'Unknown')} ({likelihood})"): st.write(f"**Rationale:** {item.get('rationale', 'N/A')}") else: st.info("No DDx provided.") st.markdown(f"**Risk Assessment:**"); risk = structured_output.get('risk_assessment', {}) flags = risk.get('identified_red_flags', []); concerns = risk.get("immediate_concerns", []); comps = risk.get("potential_complications", []) if flags: st.warning(f"**Flags:** {', '.join(flags)}") if concerns: st.warning(f"**Concerns:** {', '.join(concerns)}") if comps: st.info(f"**Potential Complications:** {', '.join(comps)}") if not flags and not concerns: st.success("No major risks highlighted.") with cols[1]: st.markdown(f"**Recommended Plan:**"); plan = structured_output.get('recommended_plan', {}) for section in ["investigations", "therapeutics", "consultations", "patient_education"]: st.markdown(f"_{section.replace('_',' ').capitalize()}:_"); items = plan.get(section) if items: [st.markdown(f"- {item}") for item in items] if isinstance(items, list) else st.markdown(f"- {items}") else: st.markdown("_None suggested._") st.markdown("") # Space st.markdown(f"**Rationale & Guideline Check:**"); st.markdown(f"> {structured_output.get('rationale_summary', 'N/A')}") interaction_summary = structured_output.get("interaction_check_summary", "") if interaction_summary: st.markdown(f"**Interaction Check Summary:**"); st.markdown(f"> {interaction_summary}") st.divider() if getattr(msg, 'tool_calls', None): with st.expander("🛠️ AI requested actions", expanded=False): for tc in msg.tool_calls: try: st.code(f"Action: {tc.get('name', 'Unknown')}\nArgs: {json.dumps(tc.get('args', {}), indent=2)}", language="json") except Exception as display_e: st.error(f"Could not display tool call: {display_e}"); st.code(str(tc)) elif isinstance(msg, ToolMessage): tool_name_display = getattr(msg, 'name', 'tool_execution') with st.chat_message(tool_name_display, avatar="🛠️", key=f"{unique_key}_tool"): # (Copy the ToolMessage display logic from previous full code) try: tool_data = json.loads(msg.content); status = tool_data.get("status", "info"); message = tool_data.get("message", msg.content) details = tool_data.get("details"); warnings = tool_data.get("warnings") if status == "success" or status == "clear" or status == "flagged": st.success(f"{message}", icon="✅" if status != "flagged" else "🚨") elif status == "warning": st.warning(f"{message}", icon="⚠️") if warnings and isinstance(warnings, list): st.caption("Details:") for warn in warnings: st.caption(f"- {warn}") # Display warnings from the tool output JSON else: st.error(f"{message}", icon="❌") if details: st.caption(f"Details: {details}") except json.JSONDecodeError: st.info(f"{msg.content}") # Display raw if not JSON except Exception as e: st.error(f"Error displaying tool message: {e}", icon="❌"); st.caption(f"Raw content: {msg.content}") # --- Chat Input Logic --- (No change needed) if prompt := st.chat_input("Your message or follow-up query..."): if not st.session_state.patient_data: st.warning("Please load patient data using the sidebar first."); st.stop() user_message = HumanMessage(content=prompt) st.session_state.messages.append(user_message) with st.chat_message("user"): st.markdown(prompt) current_state = AgentState(messages=st.session_state.messages, patient_data=st.session_state.patient_data) with st.spinner("SynapseAI is thinking..."): try: final_state = st.session_state.graph_app.invoke(current_state, {"recursion_limit": 15}) st.session_state.messages = final_state['messages'] except Exception as e: print(f"CRITICAL ERROR during graph invocation: {type(e).__name__} - {e}"); traceback.print_exc() st.error(f"An error occurred during the conversation turn: {e}", icon="❌") # Optionally add error to history for user visibility # error_ai_msg = AIMessage(content=f"Sorry, a critical error occurred: {type(e).__name__}. Please check logs or try again.") # st.session_state.messages.append(error_ai_msg) st.rerun() # Refresh display # Disclaimer (No change needed) st.markdown("---") st.warning("""**Disclaimer:** SynapseAI is an AI assistant... (Verify all outputs)""") if __name__ == "__main__": main()