SynapseAI / agent.py
mgbam's picture
Update agent.py
d21b959 verified
import os
import re
import json
import logging
import traceback
from functools import lru_cache
from typing import List, Dict, Any, Optional, TypedDict
import requests
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.pydantic_v1 import BaseModel, Field
from langchain_core.tools import tool
from langgraph.prebuilt import ToolExecutor
from langgraph.graph import StateGraph, END
# ── Logging Configuration ──────────────────────────────────────────────
logger = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO)
# ── Environment Variables ──────────────────────────────────────────────
UMLS_API_KEY = os.getenv("UMLS_API_KEY")
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
TAVILY_API_KEY = os.getenv("TAVILY_API_KEY")
if not all([UMLS_API_KEY, GROQ_API_KEY, TAVILY_API_KEY]):
logger.error("Missing one or more required API keys: UMLS_API_KEY, GROQ_API_KEY, TAVILY_API_KEY")
raise RuntimeError("Missing required API keys")
# ── Agent Configuration ──────────────────────────────────────────────
AGENT_MODEL_NAME = "llama3-70b-8192"
AGENT_TEMPERATURE = 0.1
MAX_SEARCH_RESULTS = 3
class ClinicalPrompts:
SYSTEM_PROMPT = """
You are SynapseAI, an expert AI clinical assistant engaged in an interactive consultation...
[SYSTEM PROMPT CONTENT HERE]
"""
# ── Message Normalization Helpers ─────────────────────────────────────────
def wrap_message(msg: Any) -> AIMessage:
"""
Ensures the given message is an AIMessage.
If it is a dict, extracts the 'content' field (or serializes the dict).
Otherwise, converts the message to a string.
"""
if isinstance(msg, AIMessage):
return msg
elif isinstance(msg, dict):
return AIMessage(content=msg.get("content", json.dumps(msg)))
else:
return AIMessage(content=str(msg))
def normalize_messages(state: Dict[str, Any]) -> Dict[str, Any]:
"""
Normalizes all messages in the state to be AIMessage objects.
"""
state["messages"] = [wrap_message(m) for m in state.get("messages", [])]
return state
# ── Helper Functions ─────────────────────────────────────────────────────
UMLS_AUTH_ENDPOINT = "https://utslogin.nlm.nih.gov/cas/v1/api-key"
RXNORM_API_BASE = "https://rxnav.nlm.nih.gov/REST"
OPENFDA_API_BASE = "https://api.fda.gov/drug/label.json"
@lru_cache(maxsize=256)
def get_rxcui(drug_name: str) -> Optional[str]:
"""Lookup RxNorm CUI for a given drug name."""
drug_name = (drug_name or "").strip()
if not drug_name:
return None
logger.info(f"Looking up RxCUI for '{drug_name}'")
try:
params = {"name": drug_name, "search": 1}
r = requests.get(f"{RXNORM_API_BASE}/rxcui.json", params=params, timeout=10)
r.raise_for_status()
ids = r.json().get("idGroup", {}).get("rxnormId")
if ids:
logger.info(f"Found RxCUI {ids[0]} for '{drug_name}'")
return ids[0]
r = requests.get(f"{RXNORM_API_BASE}/drugs.json", params={"name": drug_name}, timeout=10)
r.raise_for_status()
for grp in r.json().get("drugGroup", {}).get("conceptGroup", []):
props = grp.get("conceptProperties")
if props:
logger.info(f"Found RxCUI {props[0]['rxcui']} via /drugs for '{drug_name}'")
return props[0]["rxcui"]
except Exception:
logger.exception(f"Error fetching RxCUI for '{drug_name}'")
return None
@lru_cache(maxsize=128)
def get_openfda_label(rxcui: Optional[str] = None, drug_name: Optional[str] = None) -> Optional[Dict[str, Any]]:
"""Fetch the OpenFDA label for a drug by RxCUI or name."""
if not (rxcui or drug_name):
return None
terms = []
if rxcui:
terms.append(f'spl_rxnorm_code:"{rxcui}" OR openfda.rxcui:"{rxcui}"')
if drug_name:
dn = drug_name.lower()
terms.append(f'(openfda.brand_name:"{dn}" OR openfda.generic_name:"{dn}")')
query = " OR ".join(terms)
logger.info(f"Looking up OpenFDA label with query: {query}")
try:
r = requests.get(OPENFDA_API_BASE, params={"search": query, "limit": 1}, timeout=15)
r.raise_for_status()
results = r.json().get("results", [])
if results:
return results[0]
except Exception:
logger.exception("Error fetching OpenFDA label")
return None
def search_text_list(texts: List[str], terms: List[str]) -> List[str]:
"""Return highlighted snippets from a list of texts containing any of the search terms."""
snippets = []
lowers = [t.lower() for t in terms if t]
for text in texts or []:
tl = text.lower()
for term in lowers:
if term in tl:
i = tl.find(term)
start = max(0, i - 50)
end = min(len(text), i + len(term) + 100)
snippet = text[start:end]
snippet = re.sub(f"({re.escape(term)})", r"**\1**", snippet, flags=re.IGNORECASE)
snippets.append(f"...{snippet}...")
break
return snippets
def parse_bp(bp: str) -> Optional[tuple[int, int]]:
"""Parse 'SYS/DIA' blood pressure string into a (sys, dia) tuple."""
if m := re.match(r"(\d{1,3})\s*/\s*(\d{1,3})", (bp or "").strip()):
return int(m.group(1)), int(m.group(2))
return None
def check_red_flags(patient_data: Dict[str, Any]) -> List[str]:
"""Identify immediate red flags from patient_data."""
flags: List[str] = []
hpi = patient_data.get("hpi", {})
vitals = patient_data.get("vitals", {})
syms = [s.lower() for s in hpi.get("symptoms", []) if isinstance(s, str)]
mapping = {
"chest pain": "Chest pain reported",
"shortness of breath": "Shortness of breath reported",
"severe headache": "Severe headache reported",
"syncope": "Syncope reported",
"hemoptysis": "Hemoptysis reported"
}
for term, desc in mapping.items():
if term in syms:
flags.append(f"Red Flag: {desc}.")
temp = vitals.get("temp_c")
hr = vitals.get("hr_bpm")
rr = vitals.get("rr_rpm")
spo2 = vitals.get("spo2_percent")
bp = parse_bp(vitals.get("bp_mmhg", ""))
if temp is not None and temp >= 38.5:
flags.append(f"Red Flag: Fever ({temp}Β°C).")
if hr is not None:
if hr >= 120:
flags.append(f"Red Flag: Tachycardia ({hr} bpm).")
if 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:
sys, dia = bp
if sys >= 180 or dia >= 110:
flags.append(f"Red Flag: Hypertensive urgency/emergency ({sys}/{dia} mmHg).")
if sys <= 90 or dia <= 60:
flags.append(f"Red Flag: Hypotension ({sys}/{dia} mmHg).")
return list(dict.fromkeys(flags))
def format_patient_data_for_prompt(data: Dict[str, Any]) -> str:
"""Format patient_data dict into a markdown-like prompt section."""
if not data:
return "No patient data provided."
lines: List[str] = []
for section, value in data.items():
title = section.replace("_", " ").title()
if isinstance(value, dict) and any(value.values()):
lines.append(f"**{title}:**")
for k, v in value.items():
if v:
lines.append(f"- {k.replace('_',' ').title()}: {v}")
elif isinstance(value, list) and value:
lines.append(f"**{title}:** {', '.join(map(str, value))}")
elif value:
lines.append(f"**{title}:** {value}")
return "\n".join(lines)
# ── Tool Input Schemas ─────────────────────────────────────────────────────
class LabOrderInput(BaseModel):
test_name: str = Field(...)
reason: str = Field(...)
priority: str = Field("Routine")
class PrescriptionInput(BaseModel):
medication_name: str = Field(...)
dosage: str = Field(...)
route: str = Field(...)
frequency: str = Field(...)
duration: str = Field("As directed")
reason: str = Field(...)
class InteractionCheckInput(BaseModel):
potential_prescription: str
current_medications: Optional[List[str]] = Field(None)
allergies: Optional[List[str]] = Field(None)
class FlagRiskInput(BaseModel):
risk_description: str = Field(...)
urgency: str = Field("High")
# ── Tool Implementations ───────────────────────────────────────────────────
@tool("order_lab_test", args_schema=LabOrderInput)
def order_lab_test(test_name: str, reason: str, priority: str = "Routine") -> str:
"""
Place an order for a laboratory test.
"""
logger.info(f"Ordering 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:
"""
Prepare a medication prescription.
"""
logger.info(f"Preparing prescription: {medication_name} {dosage}, route: {route}, freq: {frequency}")
return json.dumps({
"status": "success",
"message": f"Prescription Prepared: {medication_name} {dosage} {route} {frequency}",
"details": f"Duration: {duration}. Reason: {reason}"
})
@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:
"""
Check for drug–drug interactions and allergy risks.
"""
logger.info(f"Checking interactions for: {potential_prescription}")
warnings: List[str] = []
pm = [m.lower().strip() for m in (current_medications or []) if m]
al = [a.lower().strip() for a in (allergies or []) if a]
if potential_prescription.lower().strip() in al:
warnings.append(f"CRITICAL ALLERGY: Patient allergic to '{potential_prescription}'.")
rxcui = get_rxcui(potential_prescription)
label = get_openfda_label(rxcui=rxcui, drug_name=potential_prescription)
if not (rxcui or label):
warnings.append(f"INFO: Could not identify '{potential_prescription}'. Checks may be incomplete.")
for section in ("contraindications", "warnings_and_cautions", "warnings"):
items = label.get(section) if label else None
if isinstance(items, list):
snippets = search_text_list(items, al)
if snippets:
warnings.append(f"ALLERGY RISK ({section}): {'; '.join(snippets)}")
for med in pm:
mrxcui = get_rxcui(med)
mlabel = get_openfda_label(rxcui=mrxcui, drug_name=med)
for sec in ("drug_interactions",):
for src_label, src_name in ((label, potential_prescription), (mlabel, med)):
items = src_label.get(sec) if src_label else None
if isinstance(items, list):
snippets = search_text_list(items, [med if src_name == potential_prescription else potential_prescription])
if snippets:
warnings.append(f"Interaction ({src_name} label): {'; '.join(snippets)}")
status = "warning" if warnings else "clear"
message = (
f"{len(warnings)} issue(s) found for '{potential_prescription}'."
if warnings else
f"No major interactions or allergy issues identified for '{potential_prescription}'."
)
return json.dumps({"status": status, "message": message, "warnings": warnings})
@tool("flag_risk", args_schema=FlagRiskInput)
def flag_risk(risk_description: str, urgency: str = "High") -> str:
"""
Flag a clinical risk with given urgency.
"""
logger.info(f"Flagging risk: {risk_description} (urgency={urgency})")
return json.dumps({
"status": "flagged",
"message": f"Risk '{risk_description}' flagged with {urgency} urgency."
})
# ── Include Tavily search tool ─────────────────────────────────────────────
search_tool = TavilySearchResults(max_results=MAX_SEARCH_RESULTS, name="tavily_search_results")
all_tools = [order_lab_test, prescribe_medication, check_drug_interactions, flag_risk, search_tool]
# ── LLM & Tool Executor ───────────────────────────────────────────────────
llm = ChatGroq(temperature=AGENT_TEMPERATURE, model=AGENT_MODEL_NAME)
model_with_tools = llm.bind_tools(all_tools)
tool_executor = ToolExecutor(all_tools)
# ── State Definition ─────────────────────────────────────────────────────
class AgentState(TypedDict):
messages: List[Any]
patient_data: Optional[Dict[str, Any]]
summary: Optional[str]
interaction_warnings: Optional[List[str]]
done: Optional[bool]
iterations: Optional[int]
# Helper to propagate state fields between nodes
def propagate_state(new: Dict[str, Any], old: Dict[str, Any]) -> Dict[str, Any]:
for key in ["iterations", "done", "patient_data", "summary", "interaction_warnings"]:
if key in old and key not in new:
new[key] = old[key]
return new
# ── Graph Nodes ─────────────────────────────────────────────────────────
def agent_node(state: AgentState) -> Dict[str, Any]:
state = normalize_messages(state)
if state.get("done", False):
return state
msgs = state.get("messages", [])
if not msgs or not isinstance(msgs[0], SystemMessage):
msgs = [SystemMessage(content=ClinicalPrompts.SYSTEM_PROMPT)] + msgs
logger.info(f"Invoking LLM with {len(msgs)} messages")
try:
response = model_with_tools.invoke(msgs)
response = wrap_message(response)
new_state = {"messages": [response]}
return propagate_state(new_state, state)
except Exception as e:
logger.exception("Error in agent_node")
new_state = {"messages": [wrap_message(AIMessage(content=f"Error: {e}"))]}
return propagate_state(new_state, state)
def tool_node(state: AgentState) -> Dict[str, Any]:
state = normalize_messages(state)
if state.get("done", False):
return state
messages_list = state.get("messages", [])
if not messages_list:
logger.warning("tool_node invoked with no messages")
new_state = {"messages": []}
return propagate_state(new_state, state)
last = wrap_message(messages_list[-1])
tool_calls = last.__dict__.get("tool_calls")
if not (isinstance(last, AIMessage) and tool_calls):
logger.warning("tool_node invoked without pending tool_calls")
new_state = {"messages": []}
return propagate_state(new_state, state)
calls = tool_calls
blocked_ids = set()
for call in calls:
if call["name"] == "prescribe_medication":
med = call["args"].get("medication_name", "").lower()
if not any(
c["name"] == "check_drug_interactions" and
c["args"].get("potential_prescription", "").lower() == med
for c in calls
):
logger.warning(f"Blocking prescribe_medication for '{med}' without interaction check")
blocked_ids.add(call["id"])
to_execute = [c for c in calls if c["id"] not in blocked_ids]
pd = state.get("patient_data", {})
for call in to_execute:
if call["name"] == "check_drug_interactions":
call["args"].setdefault("current_medications", pd.get("medications", {}).get("current", []))
call["args"].setdefault("allergies", pd.get("allergies", []))
messages: List[ToolMessage] = []
warnings: List[str] = []
try:
responses = tool_executor.batch(to_execute, return_exceptions=True)
for call, resp in zip(to_execute, responses):
if isinstance(resp, Exception):
logger.exception(f"Error executing tool {call['name']}")
content = json.dumps({"status": "error", "message": str(resp)})
else:
content = str(resp)
if call["name"] == "check_drug_interactions":
data = json.loads(content)
if data.get("status") == "warning":
warnings.extend(data.get("warnings", []))
messages.append(ToolMessage(content=content, tool_call_id=call["id"], name=call["name"]))
except Exception as e:
logger.exception("Critical error in tool_node")
for call in to_execute:
messages.append(ToolMessage(
content=json.dumps({"status": "error", "message": str(e)}),
tool_call_id=call["id"],
name=call["name"]
))
new_state = {"messages": messages, "interaction_warnings": warnings or None}
return propagate_state(new_state, state)
def reflection_node(state: AgentState) -> Dict[str, Any]:
state = normalize_messages(state)
if state.get("done", False):
return state
warns = state.get("interaction_warnings")
if not warns:
logger.warning("reflection_node called without warnings")
new_state = {"messages": []}
return propagate_state(new_state, state)
triggering = None
for msg in reversed(state.get("messages", [])):
wrapped = wrap_message(msg)
if isinstance(wrapped, AIMessage) and wrapped.__dict__.get("tool_calls"):
triggering = wrapped
break
if not triggering:
new_state = {"messages": [AIMessage(content="Internal Error: reflection context missing.")]}
return propagate_state(new_state, state)
prompt = (
"You are SynapseAI, performing a focused safety review of the following plan:\n\n"
f"{triggering.content}\n\n"
"Highlight any issues based on these warnings:\n" +
"\n".join(f"- {w}" for w in warns)
)
try:
resp = llm.invoke([SystemMessage(content="Safety reflection"), HumanMessage(content=prompt)])
new_state = {"messages": [wrap_message(resp)]}
return propagate_state(new_state, state)
except Exception as e:
logger.exception("Error during reflection")
new_state = {"messages": [AIMessage(content=f"Error during reflection: {e}")]}
return propagate_state(new_state, state)
# ── Routing Functions ────────────────────────────────────────────────────
def should_continue(state: AgentState) -> str:
state = normalize_messages(state)
state.setdefault("iterations", 0)
state["iterations"] += 1
logger.info(f"Iteration count: {state['iterations']}")
if state["iterations"] >= 4:
state.setdefault("messages", []).append(AIMessage(content="Final output: consultation complete."))
state["done"] = True
return "end_conversation_turn"
if not state.get("messages"):
state["done"] = True
return "end_conversation_turn"
last = wrap_message(state["messages"][-1])
if not isinstance(last, AIMessage):
state["done"] = True
return "end_conversation_turn"
if last.__dict__.get("tool_calls"):
return "continue_tools"
if "consultation complete" in last.content.lower():
state["done"] = True
return "end_conversation_turn"
# If no tool calls are present, terminate the conversation instead of looping.
state["done"] = True
return "end_conversation_turn"
def after_tools_router(state: AgentState) -> str:
if state.get("interaction_warnings"):
return "reflection"
return "end_conversation_turn"
# ── ClinicalAgent ─────────────────────────────────────────────────────────
class ClinicalAgent:
def __init__(self):
logger.info("Building ClinicalAgent workflow")
wf = StateGraph(AgentState)
wf.add_node("start", agent_node)
wf.add_node("tools", tool_node)
wf.add_node("reflection", reflection_node)
wf.set_entry_point("start")
wf.add_conditional_edges("start", should_continue, {
"continue_tools": "tools",
"end_conversation_turn": END
})
wf.add_conditional_edges("tools", after_tools_router, {
"reflection": "reflection",
"end_conversation_turn": END
})
self.graph_app = wf.compile()
logger.info("ClinicalAgent ready")
def invoke_turn(self, state: Dict[str, Any]) -> Dict[str, Any]:
try:
result = self.graph_app.invoke(state, {"recursion_limit": 100})
result.setdefault("summary", state.get("summary"))
result.setdefault("interaction_warnings", None)
return result
except Exception as e:
logger.exception("Error during graph invocation")
return {
"messages": state.get("messages", []) + [AIMessage(content=f"Error: {e}")],
"patient_data": state.get("patient_data"),
"summary": state.get("summary"),
"interaction_warnings": None
}