app.py

# You can find this code for Chainlit python streaming here (https://docs.chainlit.io/concepts/streaming/python)
import os
import chainlit as cl  # importing chainlit for our app
from typing import Annotated, List
from dotenv import load_dotenv
from typing_extensions import List, TypedDict

from langchain_huggingface import HuggingFaceEmbeddings
from langchain.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI
from langchain_core.documents import Document
from langchain.retrievers.contextual_compression import ContextualCompressionRetriever
from langchain_cohere import CohereRerank
from langgraph.graph import START, StateGraph, END
from langchain_core.messages import HumanMessage
from langchain_core.tools import tool
from langchain_community.tools import TavilySearchResults
from langgraph.prebuilt.tool_node import ToolNode
from langgraph.graph.message import add_messages
from langchain_community.vectorstores import FAISS
from vectorstore import VectorStore

load_dotenv()

# Using OpenAI API for embeddings/llms 
""" OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
os.environ["OPENAI_API_KEY"] = OPENAI_API_KEY

COHERE_API_KEY = os.getenv("COHERE_API_KEY")
os.environ["COHERE_API_KEY"] = COHERE_API_KEY """


# ------- Models/Tools ------- #
embed_model = HuggingFaceEmbeddings(
    model_name="Snowflake/snowflake-arctic-embed-l",
    model_kwargs={'device': 'cpu'},
    encode_kwargs={'normalize_embeddings': True}
)

llm_sml = ChatOpenAI(
        model="gpt-4o-mini", 
        temperature=0,
    )

# ------- Prompts ------- #
rag_prompt = ChatPromptTemplate.from_template("""\
    You are a helpful assistant who answers questions based on provided context. You must only use the provided context. Do NOT use your own knowledge.
    if you don't know the answer, say so.
    ### Question
    {question}
    ### Context
    {context}
    """)

# load documents and create vector store
vectorstore = VectorStore(
  collection_name="mg_alloy_collection_snowflake",
)
documents = VectorStore.load_chunks_as_documents("data/contextual_chunks")
vectorstore.add_documents(documents)
retriever = vectorstore.as_retriever(k=5)

# ------- Pydantic Models ------- #
class State(TypedDict):
  question: str
  context: List[Document]
  response: str


# ------- Functions ------- # 

def generate(state):
  docs_content = "\n\n".join(doc.page_content for doc in state["context"])
  messages = rag_prompt.format_messages(question=state["question"], context=docs_content)
  response = llm_sml.invoke(messages)
  return {"response" : response.content}


def retrieve_adjusted(state: State):
  compressor = CohereRerank(model="rerank-v3.5")
  compression_retriever = ContextualCompressionRetriever(
    base_compressor=compressor, base_retriever=retriever, search_kwargs={"k": 5}
  )
  retrieved_docs = compression_retriever.invoke(state["question"])
  return {"context" : retrieved_docs}


def should_continue(state):
  last_message = state["messages"][-1]

  if last_message.tool_calls:
    return "action"

  return END

# ------- Runnables ------- #

# retrieve graph
graph_builder = StateGraph(State)
graph_builder.add_node("retrieve", retrieve_adjusted)
graph_builder.add_node("generate", generate)
graph_builder.add_edge(START, "retrieve")
graph_builder.add_edge("retrieve", "generate")
graph_builder.add_edge("generate", END)
graph = graph_builder.compile()


@tool
def ai_rag_tool(question: str) -> str:
  """Useful for when you need to answer questions about magnesium alloys. Input should be a fully formed question."""
  response = graph.invoke({"question" : question})
  return {
        "messages": [HumanMessage(content=response["response"])],
        "context": response["context"]
    }


# ------------------------------------------------ # 
tool_belt = [
    ai_rag_tool
]


class AgentState(TypedDict):
  messages: Annotated[list, add_messages]
  context: List[Document]

tool_node = ToolNode(tool_belt)

uncompiled_graph = StateGraph(AgentState)

def call_model(state):
  messages = state["messages"]
  response = llm_sml.invoke(messages)
  return {
        "messages": [response],
        "context": state.get("context", [])
    }

uncompiled_graph.add_node("agent", call_model)
uncompiled_graph.add_node("action", tool_node)
uncompiled_graph.set_entry_point("agent")

def should_continue(state):
  last_message = state["messages"][-1]

  if last_message.tool_calls:
    return "action"

  return END

uncompiled_graph.add_conditional_edges(
    "agent",
    should_continue
)

uncompiled_graph.add_edge("action", "agent")

compiled_graph = uncompiled_graph.compile()


# ------- Chainlit ------- #
@cl.on_chat_start
async def start():
  cl.user_session.set(
      "graph", compiled_graph)

@cl.on_message
async def handle(message: cl.Message):
  graph = cl.user_session.get("graph")
  state = {"messages" : [HumanMessage(content=message.content)]}
  response = await graph.ainvoke(state)
  await cl.Message(content=response["messages"][-1].content).send()