init

apps/agent/constant.py

@@ -17,8 +17,8 @@ PROMPT = ChatPromptTemplate.from_messages(
     [
         (
             "system",
-            "You are a knowledgeable instructor. Your job is to help students learn a tool, the data for which is retrieved from a documentation site"
-            "Answer questions directly and clearly, as if you were explaining to a student who needs precise and structured guidance."
+            "You are a knowledgeable instructor. Your job is to help students learn Xano and WeWeb, the data for which is retrieved from a documentation site"
+            "Answer questions directly and clearly, as if you were explaining to students who need precise and structured guidance. Also add website sources."
             "If the answer doesn't fit the given context, just say I don't have the information for that."
         ),
         ("placeholder", "{messages}")

apps/agent/graph.py

@@ -7,7 +7,7 @@ from langchain_core.messages import AIMessage, ToolMessage
 from langgraph.prebuilt import tools_condition
 
 from langchain_groq import ChatGroq
-from apps.agent.tools import tool_weweb, tool_xano    # kalo run api pake ini -> app.agent.tools
+from apps.agent.tools import tool_weweb, tool_xano    
 from apps.agent.state import State, RequestAssistance
 from apps.agent.constant import PROMPT
 

apps/agent/multi_query_chain.py

@@ -0,0 +1,39 @@
+from typing import List
+
+from langchain_core.output_parsers import BaseOutputParser
+from langchain_core.prompts import PromptTemplate
+from pydantic import BaseModel, Field
+from langchain_groq import ChatGroq
+from apps.agent.constant import GROQ_API_KEY, MODEL_GROQ
+
+# Output parser will split the LLM result into a list of queries
+class LineListOutputParser(BaseOutputParser[List[str]]):
+    """Output parser for a list of lines."""
+
+    def parse(self, text: str) -> List[str]:
+        lines = text.strip().split("\n")
+        return list(filter(None, lines))  # Remove empty lines
+
+output_parser = LineListOutputParser()
+llm = ChatGroq(model=MODEL_GROQ, groq_api_key=GROQ_API_KEY, temperature=0.1)
+
+template = """
+Your task is to generate 3 different search queries that aim to
+answer the user question from multiple perspectives. The user questions
+are focused on Large Language Models, Machine Learning, and related
+disciplines.
+Each query MUST tackle the question from a different viewpoint, we
+want to get a variety of RELEVANT search results.
+Provide these alternative questions separated by newlines.
+GENERATE ONLY QUERY! dont add explanation and word
+Original question: {question}
+"""
+
+QUERY_PROMPT = PromptTemplate(
+    input_variables=["question"],
+    template=template,
+)
+# llm = ChatOpenAI(temperature=0, openai_api_key=OPENAI_API_KEY)
+
+# Chain
+llm_chain = QUERY_PROMPT | llm | output_parser

apps/agent/tools.py

@@ -4,34 +4,39 @@ from langchain_community.embeddings.fastembed import FastEmbedEmbeddings
 from langchain.retrievers import ContextualCompressionRetriever
 from langchain.retrievers.document_compressors import FlashrankRerank
 from langchain_core.tools import tool
+from langchain.retrievers.multi_query import MultiQueryRetriever
 
+from apps.agent.multi_query_chain import llm_chain
 from apps.agent.constant import INDEX_NAME_WEWEB, INDEX_NAME_XANO
 
-# os.environ["PINECONE_API_KEY"] = "a526d62f-ccca-40d6-859b-3d878c8d288b"
 
-embeddings = FastEmbedEmbeddings(model_name="BAAI/bge-small-en-v1.5")
-compressor = FlashrankRerank()
+embeddings = FastEmbedEmbeddings(model_name="jinaai/jina-embeddings-v2-small-en")
+compressor = FlashrankRerank(model="ms-marco-MiniLM-L-12-v2")
 
-def create_compressed_retriever(index_name: str, embeddings, compressor) -> ContextualCompressionRetriever:
+def multiquery_retriever(index_name: str, embeddings, compressor) -> ContextualCompressionRetriever:
     vectorstore = Pinecone.from_existing_index(embedding=embeddings, index_name=index_name)
     retriever = vectorstore.as_retriever()
-    return ContextualCompressionRetriever(base_compressor=compressor, base_retriever=retriever)
+    reranker_retriever = ContextualCompressionRetriever(
+        base_compressor=compressor, base_retriever=retriever
+    )
+    return MultiQueryRetriever(
+        retriever=reranker_retriever, llm_chain=llm_chain, parser_key="lines"
+    )
 
-reranker_xano = create_compressed_retriever(INDEX_NAME_XANO, embeddings, compressor)
-reranker_weweb = create_compressed_retriever(INDEX_NAME_WEWEB, embeddings, compressor)
+retriever_xano = multiquery_retriever(INDEX_NAME_XANO, embeddings, compressor)
+retriever_weweb = multiquery_retriever(INDEX_NAME_WEWEB, embeddings, compressor)
 
 @tool
 def tool_xano(query: str):
     """
     Searches and returns excerpts from the Xano documentation
     """
-    docs = reranker_xano.invoke(query)
-    return "\n\n".join([doc["page_content"] for doc in docs])
+    return retriever_xano.get_relevant_documents(query)
+    
 
 @tool
 def tool_weweb(query: str):
     """
     Searches and returns excerpts from the Weweb documentation
     """
-    docs = reranker_weweb.invoke(query)
-    return "\n\n".join([doc["page_content"] for doc in docs])
+    return retriever_weweb.get_relevant_documents(query)