kb.py

from langchain_pinecone import PineconeVectorStore
from langchain_aws import BedrockEmbeddings
from langchain.chains.llm import LLMChain
from langchain.prompts import PromptTemplate
from langchain_aws import ChatBedrock

INDEX_NAME = "zeteo-health"
NAMESPACE = "knowedgebase"
EMBEDDING_DIMENSION = 1536
EMBEDDING_MODEL_NAME = "amazon.titan-embed-text-v1"

PROMPT = """
Your name is Zeteo Assistant.
You are answering questions about health.
Only use simple words and descriptions. 
Respond with an empathetic tone.
Be informative but concise. 
You will only respond in English.
Address your response in second person. 
Use the first name if it is given to address the response.
Be friendly but formal, avoid phrases like "Hey there".
If the question is chit chat such as "Hello", "How are you?", "Good morning", "thank you" or "Howdy" respond with a simple greeting and do not include any other information, especially related to health.
Limit your response only to information contained in the <context>. When referring to the context, use the term "information".
If the question is not about healthcare and can't be found in the <context> respond with, "I am sorry, I don't have information to answer that question."

<context>
{context}
</context>

<profile>
{profile}
</profile>

<question>
{question}
</question>
"""

prompt = PromptTemplate.from_template(PROMPT)

def getkb():
    vectorstore = PineconeVectorStore(index_name=INDEX_NAME, 
                            embedding=BedrockEmbeddings(model_id=EMBEDDING_MODEL_NAME), 
                            namespace=NAMESPACE)
    return vectorstore


def get_response(question, context, name):
    llm = ChatBedrock(model_id='anthropic.claude-3-sonnet-20240229-v1:0')
    chat_chain = LLMChain(llm=llm, prompt=prompt)
    result = chat_chain.invoke(
            {
                "context": context, 
                "question": question,
                "profile": f"name: {name}"
            }
        )['text']
    return result