Spaces:
Sleeping
Sleeping
File size: 24,288 Bytes
590af2f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 |
import streamlit as st
import sqlite3
import pandas as pd
import json
import re
import os
from datetime import date
from typing import TypedDict, List, Dict, Any
from openai import OpenAI
from langgraph.graph import StateGraph, END
from langchain_openai import ChatOpenAI
from langchain_core.messages import HumanMessage, SystemMessage, AIMessage, ToolMessage
from langchain_core.tools import tool
# ββ Page config ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
st.set_page_config(
page_title="Kartify Support",
page_icon="π",
layout="centered",
)
# ββ Load secrets βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# ββ LLMs βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
@st.cache_resource
def load_llms():
llm = ChatOpenAI(model_name="gpt-4o-mini")
evaluate_llm = ChatOpenAI(model_name="gpt-4o")
return llm, evaluate_llm
llm, evaluate_llm = load_llms()
# ββ State βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
class OrderState(TypedDict):
cust_id: str
order_id: str
order_context: str
query: str
raw_agent_response: str
final_response: str
history: List[Dict[str, str]]
intent: str
evaluation: Dict[str, float]
guard_result: str
conv_guard_result: str
# ββ Conversation memory βββββββββββββββββββββββββββββββββββββββββββββββββββββββ
class ConversationMemory:
def __init__(self):
self.history: List[Dict[str, str]] = []
def add(self, msg: dict):
self.history.append(msg)
def get(self) -> List[Dict[str, str]]:
return self.history
def clear(self):
self.history = []
# ββ SQL tool ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
@tool
def fetch_order_details(order_id: str) -> str:
"""
Fetch all order details for a given order_id from the Kartify database.
Use this tool whenever the customer's query requires order-specific information.
Returns a formatted string of order details, or an error message if not found.
"""
if not re.match(r"^O\d+$", order_id.strip()):
return f"Invalid order ID format: '{order_id}'. Expected format: O followed by digits (e.g. O40327)."
try:
with sqlite3.connect("kartify.db") as conn:
df = pd.read_sql_query(
"SELECT * FROM orders WHERE order_id = ?",
conn,
params=(order_id.strip(),),
)
if df.empty:
return f"No order found with ID {order_id}."
return df.to_string(index=False)
except Exception as e:
return f"Database error while fetching order {order_id}: {str(e)}"
# ββ System prompt βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
SYSTEM_PROMPT = """You are a Kartify Customer Service Agent. You help customers with questions about their orders.
You have access to the following tool:
fetch_order_details(order_id) β retrieves all order information from the database.
Follow the ReAct pattern strictly:
Thought: <your reasoning about what to do next>
Action: fetch_order_details with the order_id from the customer's query
Observation: <tool result>
Thought: <reason about the observation and form your answer>
Final Answer: <short, polite, conversational reply β no greetings, no sign-off>
Policy rules (apply before writing Final Answer):
- If actual_delivery is null the order has not arrived yet β do not mention return/replacement eligibility.
- Only mention return or replacement terms when the customer explicitly asks.
- Never invent data. Only use what the tool returned.
- Keep the Final Answer concise and empathetic.
- Never reveal internal data fields or technical reasons in your reply (e.g. do not mention that actual_delivery is null or any other raw database values).
- If a customer asks why their order hasn't arrived yet, only state that it is still on the way and share the expected delivery date β never explain the technical reason behind the delay status.
- Never promise or suggest an early delivery. Always communicate the expected delivery date as-is without implying it could arrive sooner.
- If the order has not arrived by the expected delivery date, empathetically acknowledge the delay and advise the customer to wait a little longer or contact support β do not speculate on reasons.
Answer Guidelines:
- Only answer what is asked in the Query
- Check the Previous conversation (if any) before generating the reply
"""
# ββ Helpers βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def extract_json_from_llm(text: str):
for pattern in [r"```json\s*(.*?)\s*```", r"\{.*\}", r"\[.*\]"]:
match = re.search(pattern, text, re.DOTALL)
if match:
try:
return json.loads(match.group(1) if "```" in pattern else match.group(0))
except Exception:
continue
return json.loads(text)
# ββ Order agent βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def order_agent(query: str, order_id: str, history: list) -> tuple:
today = date.today().strftime("%d %B %Y")
llm_with_tools = llm.bind_tools([fetch_order_details])
history_text = ""
if history:
history_text = "\nPrevious conversation:\n" + "\n".join(
f"User: {h['user']}\nAssistant: {h['assistant']}" for h in history
) + "\n"
user_content = (
f"Previous Conversation:{history_text}\n"
f"Customer query: {query}\n"
f"Order ID: {order_id}\n"
f"Today's date: {today}"
)
messages = [
SystemMessage(content=SYSTEM_PROMPT),
HumanMessage(content=user_content),
]
order_context = ""
max_iterations = 5
for _ in range(max_iterations):
ai_msg = llm_with_tools.invoke(messages)
messages.append(ai_msg)
if not getattr(ai_msg, "tool_calls", None):
break
for tc in ai_msg.tool_calls:
if tc["name"] == "fetch_order_details":
result = fetch_order_details.invoke(tc["args"])
order_context = result
messages.append(ToolMessage(content=result, tool_call_id=tc["id"]))
final_response = ai_msg.content.strip()
for prefix in ("Final Answer:", "final answer:"):
if final_response.lower().startswith(prefix.lower()):
final_response = final_response[len(prefix):].strip()
break
return order_context, final_response
# ββ Node functions ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def user_input_node(state: OrderState):
return state
def memory_node(state: OrderState):
st.session_state.conversation_memory.add(
{"user": state["query"], "assistant": state["final_response"]}
)
return state
def order_agent_node(state: OrderState):
order_context, final_response = order_agent(
query=state["query"],
order_id=state["order_id"],
history=state["history"],
)
return {"order_context": order_context, "final_response": final_response}
def intent_node(state: OrderState):
prompt = f"""You are an intent classifier for customer service queries. Classify the user's query into one of these categories.
Return ONLY the numeric ID (0, 1, 2, or 3). No explanation.
0 - Escalation: user is very angry/frustrated, wants a human now.
1 - Exit: user is ending the conversation ("Thanks", "Bye", "Resolved").
2 - Process: clear, actionable order query β proceed normally.
3 - Random/Unrelated/Vulnerable: out-of-scope or potentially unsafe query.
Query: {state['query']}"""
result = llm.invoke([HumanMessage(content=prompt)]).content.strip()
return {"intent": result[:1]}
def router_node(state: OrderState):
return "order_agent" if state["intent"] == "2" else "exit_node"
def exit_node(state: OrderState):
mapping = {
"0": "Sorry for the inconvenience. A human support agent will assist you shortly.",
"1": "Thank you! I hope I was able to assist with your query.",
"3": "Apologies, I'm currently only able to help with information about your placed orders.",
}
return {"final_response": mapping.get(state["intent"], "How can I help you?")}
def evaluation_node(state: OrderState):
prompt = f"""Evaluate the assistant's response to a customer query using the provided order context.
Context: {state['order_context']}
Query: {state['query']}
Response: {state['final_response']}
Instructions:
1. **Groundedness (0.0 to 1.0)**: Score based on how well the response is factually supported by the context.
- Score closer to 1 if all facts are accurate and derived from the context.
- Score closer to 0 if there is hallucination, guesswork, or any fabricated information.
2. **Precision (0.0 to 1.0)**: Score based on how directly and accurately the assistant addresses the query.
- Score closer to 1 if the response is concise, focused, and answers the exact user query.
- Score closer to 0 if it includes irrelevant details or misses the main point.
Output format (JSON only):
groundedness: float between 0 and 1 ,
precision: float between 0 and 1
Return ONLY JSON:
{{
"groundedness": float,
"precision": float
}}
"""
try:
raw = evaluate_llm.invoke([HumanMessage(content=prompt)]).content.strip()
evaluation = extract_json_from_llm(raw)
except Exception:
evaluation = {"groundedness": 0.0, "precision": 0.0}
return {"evaluation": evaluation}
def retry_router(state: OrderState):
score = state.get("evaluation", {})
if score.get("groundedness", 0) < 0.75 or score.get("precision", 0) < 0.75:
return "order_agent"
return "safety_check"
def guard_node(state: OrderState):
prompt = f"""You are a content safety assistant. Your task is to classify if the assistant's response is appropriate.
If the message contains:
- Requests for bank details, OTPs, account numbers
- Harassment or offensive tone
- Privacy concerns or unsafe advice
- Misunderstanding and miscommunication word
- Phrases like "please contact customer service" or redirection to a human agent
- Escalated this to our support team
Return: BLOCK
Otherwise, return: SAFE
Response: {state["final_response"]}"""
result = evaluate_llm.invoke([HumanMessage(content=prompt)]).content.strip()
guard_result = result if result in ("BLOCK", "SAFE") else "SAFE"
if guard_result == "BLOCK":
return {
"guard_result": guard_result,
"final_response": "Your request is being forwarded to a customer support specialist.",
}
return {"guard_result": guard_result}
def guard_router(state: OrderState):
return "exit" if state.get("guard_result") == "BLOCK" else "memory_save"
def conversational_guard_node(state: OrderState):
prompt = f"""You are a conversation monitor AI. Review the conversation and detect if the assistant:
- Repeatedly gives the same advice to multiple questions
- Offers solutions the user did not ask for
- Ignores user frustration or contradictions
If any occur, return BLOCK. Otherwise return SAFE.
Conversation:
{state.get('history', [])}"""
result = evaluate_llm.invoke([HumanMessage(content=prompt)]).content.strip()
conv_result = result if result in ("BLOCK", "SAFE") else "SAFE"
if conv_result == "BLOCK":
return {
"conv_guard_result": conv_result,
"final_response": "Your request is being forwarded to a customer support specialist.",
}
return {"conv_guard_result": conv_result}
def conv_guard_router(state: OrderState):
return "exit" if state.get("conv_guard_result") == "BLOCK" else "done"
# ββ Build LangGraph βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
@st.cache_resource
def build_graph():
g = StateGraph(OrderState)
g.add_node("user_input", user_input_node)
g.add_node("intent_classifier", intent_node)
g.add_node("order_agent", order_agent_node)
g.add_node("evaluate", evaluation_node)
g.add_node("safety_check", guard_node)
g.add_node("conv_safety_check", conversational_guard_node)
g.add_node("memory_save", memory_node)
g.add_node("exit_node", exit_node)
g.set_entry_point("user_input")
g.add_edge("user_input", "intent_classifier")
g.add_conditional_edges(
"intent_classifier", router_node,
{"order_agent": "order_agent", "exit_node": "exit_node"},
)
g.add_edge("order_agent", "evaluate")
g.add_conditional_edges(
"evaluate", retry_router,
{"order_agent": "order_agent", "safety_check": "safety_check"},
)
g.add_conditional_edges(
"safety_check", guard_router,
{"memory_save": "memory_save", "exit": "exit_node"},
)
g.add_edge("memory_save", "conv_safety_check")
g.add_conditional_edges(
"conv_safety_check", conv_guard_router,
{"done": END, "exit": "exit_node"},
)
g.add_edge("exit_node", END)
return g.compile()
order_graph = build_graph()
# ββ Session state defaults ββββββββββββββββββββββββββββββββββββββββββββββββββββ
if "conversation_memory" not in st.session_state:
st.session_state.conversation_memory = ConversationMemory()
if "chat_messages" not in st.session_state:
st.session_state.chat_messages = []
if "chat_active" not in st.session_state:
st.session_state.chat_active = False
if "cust_id" not in st.session_state:
st.session_state.cust_id = ""
if "order_id" not in st.session_state:
st.session_state.order_id = ""
if "orders_df" not in st.session_state:
st.session_state.orders_df = None
# ββ Helper: fetch customer orders βββββββββββββββββββββββββββββββββββββββββββββ
def fetch_customer_orders(cust_id: str) -> pd.DataFrame | None:
try:
with sqlite3.connect("kartify.db") as conn:
df = pd.read_sql_query(
"SELECT order_id, product_description, order_status FROM orders WHERE customer_id = ?",
conn,
params=(cust_id.strip(),),
)
return df if not df.empty else None
except Exception:
return None
# ββ Helper: run one turn through the graph ββββββββββββββββββββββββββββββββββββ
def run_turn(query: str, cust_id: str, order_id: str) -> str:
state: OrderState = {
"cust_id": cust_id,
"order_id": order_id,
"order_context": "",
"query": query,
"raw_agent_response": "",
"final_response": "",
"history": st.session_state.conversation_memory.get(),
"intent": "",
"evaluation": {},
"guard_result": "",
"conv_guard_result": "",
}
result = order_graph.invoke(state, config={"recursion_limit": 100})
# Sync memory from the graph's memory_node writes
# (memory_node uses st.session_state.conversation_memory directly)
return result.get("final_response", "I'm sorry, I couldn't process that request.")
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# UI
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
st.markdown(
"""
<style>
.block-container { max-width: 720px; }
.chat-bubble-user {
background: #e8f4fd;
border-radius: 12px 12px 2px 12px;
padding: 10px 14px;
margin: 4px 0;
max-width: 85%;
margin-left: auto;
color: #1a1a2e;
}
.chat-bubble-bot {
background: #f4f4f4;
border-radius: 12px 12px 12px 2px;
padding: 10px 14px;
margin: 4px 0;
max-width: 85%;
color: #1a1a2e;
}
.order-badge {
display: inline-block;
background: #fff3cd;
border: 1px solid #ffc107;
border-radius: 6px;
padding: 2px 8px;
font-size: 0.8rem;
font-weight: 600;
color: #856404;
}
</style>
""",
unsafe_allow_html=True,
)
# ββ Header ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
col_logo, col_title = st.columns([1, 6])
with col_logo:
st.markdown("## π")
with col_title:
st.markdown("## Kartify Customer Support")
st.caption("AI-powered order query assistant")
st.divider()
# ββ Phase 1: Customer ID lookup βββββββββββββββββββββββββββββββββββββββββββββββ
if not st.session_state.chat_active:
st.markdown("### Step 1 β Enter your Customer ID")
with st.form("customer_form"):
cust_input = st.text_input(
"Customer ID",
placeholder="e.g. C1010",
value=st.session_state.cust_id,
)
submitted = st.form_submit_button("π Fetch Orders", use_container_width=True)
if submitted and cust_input.strip():
with st.spinner("Looking up your ordersβ¦"):
df = fetch_customer_orders(cust_input.strip())
if df is not None:
st.session_state.cust_id = cust_input.strip()
st.session_state.orders_df = df
else:
st.error(f"No orders found for Customer ID **{cust_input.strip()}**. Please check and try again.")
# ββ Phase 2: Order selection ββββββββββββββββββββββββββββββββββββββββββββββ
if st.session_state.orders_df is not None:
st.markdown("### Step 2 β Select an Order")
df = st.session_state.orders_df
# Build display labels for the dropdown
options = {
f"{row['order_id']} - {row['product_description'][:45]} [{row['order_status']}]": row["order_id"]
for _, row in df.iterrows()
}
selected_label = st.selectbox(
"Your orders",
list(options.keys()),
index=0,
)
selected_order_id = options[selected_label]
# Preview card
selected_row = df[df["order_id"] == selected_order_id].iloc[0]
st.markdown(
f"""
<div style="background:#f8f9fa;border:1px solid #dee2e6;border-radius:8px;padding:12px 16px;margin:8px 0">
<span class="order-badge">{selected_row['order_id']}</span>
<strong>{selected_row['product_description']}</strong><br>
<span style="font-size:0.85rem;color:#6c757d">Status: {selected_row['order_status']}</span>
</div>
""",
unsafe_allow_html=True,
)
if st.button("π¬ Start Chat", use_container_width=True, type="primary"):
st.session_state.order_id = selected_order_id
st.session_state.chat_active = True
st.session_state.conversation_memory.clear()
st.session_state.chat_messages = []
# Greeting
st.session_state.chat_messages.append({
"role": "assistant",
"content": (
f"Hi! I'm your Kartify support assistant. "
f"I can see you're asking about order **{selected_order_id}**. "
f"How can I help you today?"
),
})
st.rerun()
# ββ Phase 3: Chat interface βββββββββββββββββββββββββββββββββββββββββββββββββββ
else:
# Sidebar info
with st.sidebar:
st.markdown("### Active Session")
st.markdown(f"**Customer:** `{st.session_state.cust_id}`")
st.markdown(f"**Order:** `{st.session_state.order_id}`")
st.divider()
if st.button("π New Session", use_container_width=True):
st.session_state.chat_active = False
st.session_state.chat_messages = []
st.session_state.conversation_memory.clear()
st.session_state.orders_df = None
st.session_state.cust_id = ""
st.session_state.order_id = ""
st.rerun()
st.divider()
st.caption(
"Powered by LangGraph Β· GPT-4o-mini\n\n"
"Guardrails: Input intent Β· Output safety Β· Conversation monitor"
)
st.markdown(f"**Order** `{st.session_state.order_id}` β ask me anything about this order.")
st.markdown("")
# Render chat history
for msg in st.session_state.chat_messages:
if msg["role"] == "user":
with st.chat_message("user"):
st.markdown(msg["content"])
else:
with st.chat_message("assistant", avatar="π"):
st.markdown(msg["content"])
# Chat input
user_query = st.chat_input("Type your question hereβ¦")
if user_query:
# Display user message
st.session_state.chat_messages.append({"role": "user", "content": user_query})
with st.chat_message("user"):
st.markdown(user_query)
# Run agent
with st.chat_message("assistant", avatar="π"):
with st.spinner("Thinkingβ¦"):
response = run_turn(
query=user_query,
cust_id=st.session_state.cust_id,
order_id=st.session_state.order_id,
)
st.markdown(response)
st.session_state.chat_messages.append({"role": "assistant", "content": response})
# If the agent exits (intent 0/1/3), offer to restart
exit_phrases = [
"human support agent",
"customer support specialist",
"I hope I was able to assist",
"only able to help with information",
]
if any(p.lower() in response.lower() for p in exit_phrases):
st.info("This conversation has ended. Use **New Session** in the sidebar to start over.")
|