chore: first draft of financial-reconciliation-agent

samples/financial-reconciliation-agent/agent_graph.py

@@ -0,0 +1,183 @@
+import enum
+from contextlib import asynccontextmanager
+from typing import Literal, TypedDict, Any, Optional
+
+from langchain_anthropic import ChatAnthropic
+from langgraph.graph import START, StateGraph, END
+from langgraph.prebuilt import create_react_agent
+from langgraph.types import Command
+from pydantic import BaseModel
+from uipath_langchain.chat.models import UiPathAzureChatOpenAI
+from langchain_mcp_adapters.client import MultiServerMCPClient
+
+from prompts import email_triage_prompt, refund_payment_agent_prompt, email_topic_extractor_prompt
+from local_tools import retrieve_from_execution_context_tool
+import os
+
+
+class EmailTopic(enum.Enum):
+    """Email regarding a new payment"""
+    PAYMENT = "PAYMENT"
+    """Email regarding a payment refund"""
+    REFUND = "REFUND"
+    """Used for any email topic, other than payment or refund"""
+    OTHER = "OTHER"
+
+class EmailTopicExtractorStructure(TypedDict):
+    """The structure for email topic extraction response"""
+    email_topic: str
+
+if os.getenv("USE_UIPATH_AI_UNITS") and os.getenv("USE_UIPATH_AI_UNITS") == "true":
+    # other available UiPath chat models
+    # "anthropic.claude-3-5-sonnet-20240620-v1:0",
+    # "anthropic.claude-3-5-sonnet-20241022-v2:0",
+    # "anthropic.claude-3-7-sonnet-20250219-v1:0",
+    # "anthropic.claude-3-haiku-20240307-v1:0",
+    # "gemini-1.5-pro-001",
+    # "gemini-2.0-flash-001",
+    # "gpt-4o-2024-05-13",
+    # "gpt-4o-2024-08-06",
+    # "gpt-4o-2024-11-20",
+    # "gpt-4o-mini-2024-07-18",
+    # "o3-mini-2025-01-31",
+    llm = UiPathAzureChatOpenAI()
+else:
+    llm = ChatAnthropic(model="claude-3-5-sonnet-latest")
+
+class OutputStructure(TypedDict):
+    """LLM message after finishing execution"""
+    message: str
+    """Whether agent execution should continue"""
+    should_continue: bool
+
+class GraphInput(BaseModel):
+    email_address: str
+    email_content: str
+
+class GraphOutput(BaseModel):
+    answer: str
+
+class State(BaseModel):
+    email_address: str
+    email_content: str
+    agent_message: str
+    email_topic: Optional[EmailTopic]
+    should_continue: bool
+
+def prepare_input(state: GraphInput):
+    return State(
+        email_address=state.email_address,
+        email_content=state.email_content,
+        should_continue=True,
+        agent_message="",
+        email_topic=None,
+    )
+
+@asynccontextmanager
+async def agent_mcp(
+        server_slug: str,
+        structured_output: Any = None,
+        extra_tools: Any = None):
+    async with MultiServerMCPClient() as client:
+        await client.connect_to_server_via_sse(
+            server_name="local-stripe-server",
+            url=server_slug,
+            headers={
+                "Authorization": f"Bearer {os.getenv('UIPATH_ACCESS_TOKEN')}"
+                },
+            timeout=60,
+        )
+
+        mcp_tools = client.get_tools()
+        if extra_tools:
+            available_tools = [*mcp_tools, *extra_tools]
+        else:
+            available_tools = [*mcp_tools]
+        if structured_output is not None:
+            agent = create_react_agent(llm, tools=available_tools, response_format=structured_output)
+        else:
+            agent = create_react_agent(llm, tools=available_tools)
+
+        try:
+            yield agent
+        finally:
+            pass
+
+async def understand_email(state: State) -> Command:
+    result = await llm.with_structured_output(EmailTopicExtractorStructure).ainvoke(
+            [("system", email_topic_extractor_prompt),
+             ("user", "email content: " + state.email_content)]
+        )
+    print(result)
+    return Command(
+        update={
+            "email_topic":result["email_topic"]
+        }
+    )
+
+async def triage_email(state: State) -> Command:
+    async with (agent_mcp(
+            os.getenv("UIPATH_MCP_INTERNAL_SERVER_URL"),
+            structured_output = OutputStructure,
+            extra_tools = [retrieve_from_execution_context_tool])
+    as agent):
+        response = await agent.ainvoke(
+            {
+                "messages":[("system", email_triage_prompt),
+                            ("user", "email topic: " + str(state.email_topic.value)),
+                            ("user", "email address: " + state.email_address)]
+            }
+        )
+        # Extract the message from the agent's response
+        output = response["structured_response"]
+        print(output)
+        return Command(
+            update={
+                "agent_message": output["message"],
+                "should_continue": output["should_continue"],
+            }
+        )
+
+async def handle_transaction(state: State) -> Command:
+    async with agent_mcp(os.getenv("UIPATH_MCP_EXTERNAL_SERVER_URL")) as agent:
+        response = await agent.ainvoke(
+            {
+                "messages":[("system", refund_payment_agent_prompt),
+                            ("user", "email content:" + state.email_content),
+                            ("user", "email address:" + state.email_address),]
+            }
+        )
+    return Command(
+        update={
+            "agent_message": str(response["messages"][-1].content),
+        }
+    )
+def collect_output(state: State) -> GraphOutput:
+    return GraphOutput(answer=str(state.agent_message))
+
+def decide_next_node_after_email_validation(state: State) -> Literal["handle_transaction", "collect_output"]:
+    if state.should_continue:
+        return "handle_transaction"
+    return "collect_output"
+
+def decide_next_node_given_email_topic(state: State) -> Literal["collect_output", "triage_email"]:
+    if state.email_topic == EmailTopic.OTHER:
+        return "collect_output"
+    return "triage_email"
+
+builder = StateGraph(State, input=GraphInput, output=GraphOutput)
+builder.add_node("prepare_input", prepare_input)
+builder.add_node("triage_email", triage_email)
+builder.add_node("collect_output", collect_output)
+builder.add_node("handle_transaction", handle_transaction)
+builder.add_node("understand_email", understand_email)
+
+builder.add_edge(START, "prepare_input")
+builder.add_edge("prepare_input", "understand_email")
+builder.add_conditional_edges("understand_email", decide_next_node_given_email_topic)
+builder.add_conditional_edges("triage_email", decide_next_node_after_email_validation)
+builder.add_edge("handle_transaction", "collect_output")
+builder.add_edge("collect_output", END)
+
+
+graph = builder.compile()

samples/financial-reconciliation-agent/input.json

@@ -0,0 +1,4 @@
+{
+  "email_address": "[email protected]",
+  "email_content": "I want a refund for my latest transaction of cat food"
+}

samples/financial-reconciliation-agent/langgraph.json

@@ -0,0 +1,7 @@
+{
+  "dependencies": ["."],
+  "graphs": {
+    "agent": "./agent_graph.py:graph"
+  },
+  "env": ".env"
+}

samples/financial-reconciliation-agent/local_tools.py

@@ -0,0 +1,22 @@
+from langchain_core.tools import Tool
+
+def retrieve_from_execution_context(key: str) -> str:
+    import os
+    if os.getenv(key) is None:
+        return 'not found'
+    else:
+        return os.getenv(key)
+
+
+retrieve_from_execution_context_tool = Tool.from_function(
+    func=retrieve_from_execution_context,
+    name="retrieve_from_execution_context",
+    description=""" Retrieve an execution context detail
+
+    Args:
+        key (str): The key of the element to return
+
+    Returns:
+        str: The value of the element in the execution context if found, else 'not found'
+    """,
+)

samples/financial-reconciliation-agent/mcp-internal/mcp.json

@@ -0,0 +1,9 @@
+{
+  "servers": {
+    "local-stripe-server": {
+        "transport": "stdio",
+        "command": "python",
+        "args": ["server.py"]
+    }
+  }
+}

samples/financial-reconciliation-agent/mcp-internal/pyproject.toml

@@ -0,0 +1,10 @@
+[project]
+name = "local-financial-reconciliation-mcp-server"
+version = "0.0.1"
+description = "Local MCP server for stripe agent"
+authors = [{ name = "Radu Mocanu" }]
+dependencies = [
+    "uipath-mcp>=0.0.84",
+    "uipath==2.0.56"
+]
+requires-python = ">=3.10"