Email LLM/LangChain Example

email_answer.py

# pip install -q langchain openai chromadb tiktoken
import os

from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.vectorstores import Chroma
from langchain.text_splitter import CharacterTextSplitter
from langchain.llms import OpenAI
from langchain.chains import RetrievalQA, LLMChain, TransformChain, SimpleSequentialChain, SequentialChain
from langchain.document_loaders import TextLoader
from langchain.prompts import PromptTemplate
from langchain.tools.zapier.tool import ZapierNLARunAction
from langchain.utilities.zapier import ZapierNLAWrapper

# get from https://platform.openai.com/
os.environ["OPENAI_API_KEY"] = "<OPENAI_API_KEY>"

# get from https://nla.zapier.com/demo/provider/debug (under User Information, after logging in): 
os.environ["ZAPIER_NLA_API_KEY"] = "<ZAPIER_NLA_API_KEY>"

loader = TextLoader("/content/resume.txt")
documents = loader.load()

text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
texts = text_splitter.split_documents(documents)

embeddings = OpenAIEmbeddings()

docsearch = Chroma.from_documents(texts, embeddings)

actions = ZapierNLAWrapper().list()


## step 1. gmail find email

GMAIL_SEARCH_INSTRUCTIONS = "Grab the latest email from David Ian"

def nla_gmail(inputs):
    action = next((a for a in actions if a["description"].startswith("Gmail: Find Email")), None)
    return {"email_data": ZapierNLARunAction(action_id=action["id"], zapier_description=action["description"], params_schema=action["params"]).run(inputs["instructions"])}
gmail_chain = TransformChain(input_variables=["instructions"], output_variables=["email_data"], transform=nla_gmail)



## step 2. generate retrieval question

question_template = """Converts this email into a one sentence question. Be concise. Use less than 30 words. Output question in plain text (not JSON).

Incoming email:
{email_data}

Question:"""

prompt_template = PromptTemplate(input_variables=["email_data"], template=question_template)
question_chain = LLMChain(llm=OpenAI(temperature=.7), prompt=prompt_template)


## step 3. draft email from context

context_template = """Use the following pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up an answer.
Use the answer to write an email.

{context}

Question: {question}
Draft email:"""
PROMPT = PromptTemplate(
    template=context_template, input_variables=["context", "question"]
)
chain_type_kwargs = {"prompt": PROMPT}

qa = RetrievalQA.from_chain_type(llm=OpenAI(), chain_type="stuff", retriever=docsearch.as_retriever(), chain_type_kwargs=chain_type_kwargs)

retrieval = lambda inputs: {"retrieved_data": qa.run(inputs["question"])}

retrieval_chain = TransformChain(input_variables=["question"], output_variables=["retrieved_data"], transform=retrieval)


## step 4. save drafted email

def create_draft(inputs):
    action = next((a for a in actions if a["description"].startswith("Gmail: Create Draft")), None)
    instructions = f'Create a draft with this: {inputs["draft_reply"]}'
    return {"gmail_data": ZapierNLARunAction(action_id=action["id"], zapier_description=action["description"], params_schema=action["params"]).run(instructions)}

draft_chain = TransformChain(input_variables=["draft_reply"], output_variables=["gmail_data"], transform=create_draft)

## finally, execute

overall_chain = SimpleSequentialChain(chains=[gmail_chain, question_chain, retrieval_chain, draft_chain], verbose=True)
overall_chain.run(GMAIL_SEARCH_INSTRUCTIONS)