Merge pull request #27 from dkimds/main

통신을 GET에서 POST로 변경

Author: dkimds

.gitignore

@@ -11,3 +11,4 @@ mlartifacts
 mlruns
 *.ipynb
 /.idea
+vllm/

chains/__init__.py

@@ -0,0 +1,19 @@
+"""
+AI Chain modules for different conversation types and routing.
+"""
+
+from .classification import ClassificationChain
+from .news import NewsChain
+from .finance import FinanceChain
+from .general import GeneralChain
+from .reset import ResetChain
+from .router import ChainRouter
+
+__all__ = [
+    'ClassificationChain',
+    'NewsChain', 
+    'FinanceChain',
+    'GeneralChain',
+    'ResetChain',
+    'ChainRouter'
+]

chains/classification.py

@@ -0,0 +1,96 @@
+"""
+Classification chain for determining the type of user query.
+"""
+
+from langchain_openai import ChatOpenAI
+from langchain_core.output_parsers import StrOutputParser
+from langchain_core.prompts import PromptTemplate
+from langchain_core.runnables import RunnableLambda
+from typing import Dict, Any
+
+
+class ClassificationChain:
+    """
+    Classifies user questions into categories: 최신소식, 전문지식, 리셋, or 기타
+    """
+    
+    def __init__(self, model: str = "gpt-4o-mini"):
+        """
+        Initialize the classification chain.
+        
+        Args:
+            model: OpenAI model to use for classification
+        """
+        self.model = model
+        self._setup_chain()
+    
+    def _setup_chain(self):
+        """Set up the classification chain with prompt and model."""
+        
+        # Classification prompt template
+        self.prompt = PromptTemplate.from_template(
+            """주어진 사용자 질문과 대화 히스토리를 보고 `최신소식`, `전문지식`, `리셋`, 또는 `기타` 중 하나로 분류하세요. 
+리셋 관련 키워드: "리셋", "초기화", "지워", "새로시작", "reset", "clear" 등
+한 단어 이상으로 응답하지 마세요.
+
+<chat_history>
+{chat_history}
+</chat_history>
+
+<question>
+{question}
+</question>
+
+Classification:"""
+        )
+        
+        # Create the chain
+        self.chain = (
+            self.prompt
+            | ChatOpenAI(model=self.model)
+            | StrOutputParser()
+        )
+    
+    def classify(self, question: str, chat_history: str = "") -> str:
+        """
+        Classify a user question.
+        
+        Args:
+            question: The user's question
+            chat_history: Previous conversation history
+            
+        Returns:
+            Classification result as string
+        """
+        return self.chain.invoke({
+            "question": question,
+            "chat_history": chat_history
+        })
+    
+    def invoke(self, inputs: Dict[str, Any]) -> str:
+        """
+        Invoke the classification chain with inputs.
+        
+        Args:
+            inputs: Dictionary containing 'question' and 'chat_history'
+            
+        Returns:
+            Classification result as string
+        """
+        return self.chain.invoke(inputs)
+    
+    async def ainvoke(self, inputs: Dict[str, Any]) -> str:
+        """
+        Asynchronously invoke the classification chain.
+        
+        Args:
+            inputs: Dictionary containing 'question' and 'chat_history'
+            
+        Returns:
+            Classification result as string
+        """
+        return await self.chain.ainvoke(inputs)
+    
+    def get_chain(self):
+        """Get the underlying chain object."""
+        return self.chain

chains/finance.py

@@ -0,0 +1,162 @@
+"""
+Finance chain for handling financial expert knowledge queries.
+"""
+
+from langchain_openai import ChatOpenAI, OpenAIEmbeddings
+from langchain_core.output_parsers import StrOutputParser
+from langchain_core.prompts import PromptTemplate
+from langchain_core.runnables import RunnablePassthrough, RunnableLambda
+from langchain_milvus import Milvus
+from typing import Dict, Any, Optional
+import os
+
+
+class FinanceChain:
+    """
+    Handles queries about financial expert knowledge using vector database search.
+    """
+    
+    def __init__(self, model: str = "gpt-4o-mini", search_k: int = 3):
+        """
+        Initialize the finance chain.
+        
+        Args:
+            model: OpenAI model to use for response generation
+            search_k: Number of search results to retrieve from vector DB
+        """
+        self.model = model
+        self.search_k = search_k
+        self.use_milvus = False
+        self.vectorstore = None
+        self._setup_milvus()
+        self._setup_chain()
+    
+    def _setup_milvus(self):
+        """Set up Milvus vector database connection."""
+        try:
+            embeddings = OpenAIEmbeddings()
+            milvus_uri = os.getenv("MILVUS_URI", "http://localhost:19530")
+            print(f"Milvus 연결 시도: {milvus_uri}")
+            
+            self.vectorstore = Milvus(
+                embedding_function=embeddings,
+                connection_args={
+                    "uri": milvus_uri,
+                },
+                collection_name="coindesk_articles",
+            )
+            print("Milvus 연결 성공")
+            self.use_milvus = True
+        except Exception as e:
+            print(f"Milvus 연결 실패: {e}")
+            print("주의: Milvus 연결 실패로 벡터 검색 기능을 사용할 수 없습니다.")
+            self.use_milvus = False
+            self.vectorstore = None
+    
+    def _setup_chain(self):
+        """Set up the finance chain with retriever, prompt and model."""
+        
+        # Finance prompt template
+        self.prompt = PromptTemplate(
+            template="""You are an expert in finance. \
+Always answer questions starting with "전문가에 따르면..". \
+
+Previous conversation:
+{chat_history}
+
+Respond to the following question based the context, statistical information, and previous conversation when possible:
+Context: {context}
+Question: {question}
+Answer:""",
+            input_variables=["context", "question", "chat_history"]
+        )
+        
+        # Create the chain based on Milvus availability
+        if self.use_milvus and self.vectorstore:
+            self.retriever = self.vectorstore.as_retriever(search_kwargs={"k": self.search_k})
+            context_chain = RunnableLambda(lambda x: x["question"]) | self.retriever | self._format_docs
+            
+            self.chain = (
+                {
+                    "context": context_chain,
+                    "question": RunnablePassthrough(),
+                    "chat_history": lambda x: x.get("chat_history", "")
+                }
+                | self.prompt
+                | ChatOpenAI(model=self.model)
+                | StrOutputParser()
+            )
+        else:
+            # Fallback chain when Milvus is not available
+            fallback_prompt = PromptTemplate.from_template(
+                """전문가에 따르면, 현재 벡터 데이터베이스에 연결할 수 없어 전문 지식을 제공하기 어렵습니다. 
+시스템 관리자에게 문의해주세요.
+
+Previous conversation:
+{chat_history}
+
+Question: {question}
+"""
+            )
+            
+            self.chain = (
+                fallback_prompt
+                | ChatOpenAI(model=self.model)
+                | StrOutputParser()
+            )
+    
+    def _format_docs(self, docs):
+        """Format retrieved documents for use in the prompt."""
+        return "\n\n".join(doc.page_content for doc in docs)
+    
+    def process(self, question: str, chat_history: str = "") -> str:
+        """
+        Process a finance-related question.
+        
+        Args:
+            question: The user's question
+            chat_history: Previous conversation history
+            
+        Returns:
+            Response based on financial expert knowledge
+        """
+        return self.chain.invoke({
+            "question": question,
+            "chat_history": chat_history
+        })
+    
+    def invoke(self, inputs: Dict[str, Any]) -> str:
+        """
+        Invoke the finance chain with inputs.
+        
+        Args:
+            inputs: Dictionary containing 'question' and optional 'chat_history'
+            
+        Returns:
+            Finance-based response as string
+        """
+        return self.chain.invoke(inputs)
+    
+    async def ainvoke(self, inputs: Dict[str, Any]) -> str:
+        """
+        Asynchronously invoke the finance chain.
+        
+        Args:
+            inputs: Dictionary containing 'question' and optional 'chat_history'
+            
+        Returns:
+            Finance-based response as string
+        """
+        return await self.chain.ainvoke(inputs)
+    
+    def get_chain(self):
+        """Get the underlying chain object."""
+        return self.chain
+    
+    def get_retriever(self) -> Optional[Any]:
+        """Get the vector database retriever if available."""
+        return self.retriever if self.use_milvus else None
+    
+    def is_milvus_available(self) -> bool:
+        """Check if Milvus is available for use."""
+        return self.use_milvus

chains/general.py

@@ -0,0 +1,94 @@
+"""
+General chain for handling miscellaneous queries.
+"""
+
+from langchain_openai import ChatOpenAI
+from langchain_core.output_parsers import StrOutputParser
+from langchain_core.prompts import PromptTemplate
+from typing import Dict, Any
+
+
+class GeneralChain:
+    """
+    Handles general queries that don't fit into specific categories.
+    """
+    
+    def __init__(self, model: str = "gpt-4o-mini"):
+        """
+        Initialize the general chain.
+        
+        Args:
+            model: OpenAI model to use for response generation
+        """
+        self.model = model
+        self._setup_chain()
+    
+    def _setup_chain(self):
+        """Set up the general chain with prompt and model."""
+        
+        # General prompt template
+        self.prompt = PromptTemplate.from_template(
+            """Previous conversation:
+{chat_history}
+
+Respond to the following question concisely:
+If the question is not about expert knowledge or recent events, reply:
+
+"도와드리지 못해서 죄송합니다. 저는 비트코인 관련 전문지식과 최신소식만 답변드릴 수 있습니다."
+
+Only respond with factual, concise answers supported by the context when applicable.
+Question: {question}
+Answer:
+"""
+        )
+        
+        # Create the chain
+        self.chain = (
+            self.prompt
+            | ChatOpenAI(model=self.model)
+            | StrOutputParser()
+        )
+    
+    def process(self, question: str, chat_history: str = "") -> str:
+        """
+        Process a general question.
+        
+        Args:
+            question: The user's question
+            chat_history: Previous conversation history
+            
+        Returns:
+            General response or polite refusal
+        """
+        return self.chain.invoke({
+            "question": question,
+            "chat_history": chat_history
+        })
+    
+    def invoke(self, inputs: Dict[str, Any]) -> str:
+        """
+        Invoke the general chain with inputs.
+        
+        Args:
+            inputs: Dictionary containing 'question' and optional 'chat_history'
+            
+        Returns:
+            General response as string
+        """
+        return self.chain.invoke(inputs)
+    
+    async def ainvoke(self, inputs: Dict[str, Any]) -> str:
+        """
+        Asynchronously invoke the general chain.
+        
+        Args:
+            inputs: Dictionary containing 'question' and optional 'chat_history'
+            
+        Returns:
+            General response as string
+        """
+        return await self.chain.ainvoke(inputs)
+    
+    def get_chain(self):
+        """Get the underlying chain object."""
+        return self.chain