Valentine’s gift idea using RAG: Roided out LLMs

Its Valentine’s day again. (The bane of a lover’s existence)

For on this fated day, the perfect gift must be chosen for your lover else you might incur a fury like Hell hath none. 👿👿👿

You browse some sites for quick gifting ideas but nothing seems to be good enough!

You zero in on a couple of options. Will she like it?

Will she or won’t she? 😓😓😓

Why kill yourself? Make the easy choice. Teach an LLM to decide the idea for you.

I browsed the internet for some sites that have good gifting ideas. Then I shamelessly pulled the HTML content, scrubbed it for my intents and purposes and shoved them into PDF files.

Hint: BeautifulSoup

Now that I have the PDF files, what next?

Well here’s a recipe to train your favorite model to generate gifting ideas

𝗦𝘁𝗲𝗽 𝟭 : 𝗔𝗱𝗱 𝗮 𝗯𝘂𝗻𝗰𝗵 𝗼𝗳 𝗟𝗮𝗻𝗴𝗰𝗵𝗮𝗶𝗻 𝗱𝗲𝗽𝗲𝗻𝗱𝗲𝗻𝗰𝗶𝗲𝘀

import os
from langchain.document_loaders import PyPDFDirectoryLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.retrievers.self_query.base import SelfQueryRetriever
from langchain.vectorstores import Chroma
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.llms import OpenAI
from langchain.chains.query_constructor.base import AttributeInfo
from langchain.chat_models import ChatOpenAI
from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain

𝗦𝘁𝗲𝗽 𝟮 : 𝗚𝗲𝘁 𝗮𝗻 𝗢𝗽𝗲𝗻 𝗔𝗜 𝗸𝗲𝘆 𝗮𝗻𝗱 𝗹𝗼𝗮𝗱 𝗶𝘁 𝗶𝗻𝘁𝗼 𝘁𝗵𝗲 𝗲𝗻𝘃

with open("./openai-key.txt") as oakf:
    os.environ["OPENAI_API_KEY"] = oakf.read()

𝗦𝘁𝗲𝗽 𝟯 : 𝗟𝗼𝗮𝗱 𝘁𝗵𝗲 𝗽𝗱𝗳𝘀

loader = PyPDFDirectoryLoader("data")
data = loader.load()

𝗦𝘁𝗲𝗽 𝟰 : 𝗦𝗽𝗹𝗶𝘁 𝘁𝗵𝗲 𝗰𝗼𝗻𝘁𝗲𝗻𝘁 𝘂𝘀𝗶𝗻𝗴 𝗥𝗲𝗰𝘂𝗿𝘀𝗶𝘃𝗲𝗧𝗲𝘅𝘁𝗦𝗽𝗹𝗶𝘁𝘁𝗲𝗿

r_splitter = RecursiveCharacterTextSplitter(
    chunk_size=450,
    chunk_overlap=0,
    separators=["\n\n", "\n", " "]
)

splits = r_splitter.split_documents(data)
print(splits[0])

𝗦𝘁𝗲𝗽 𝟱 : 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗲 𝗲𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀 𝘁𝗼 𝗱𝘂𝗺𝗽 𝗶𝗻𝘁𝗼 𝗮 𝗩𝗲𝗰𝘁𝗼𝗿 𝗱𝗮𝘁𝗮𝗯𝗮𝘀𝗲 (𝗖𝗵𝗿𝗼𝗺𝗮 𝗗𝗕)

embedding = OpenAIEmbeddings()

# save chroma db embeddings in this directory
persist_directory = 'docs/chroma/'

# Create the vector store
vectordb = Chroma.from_documents(
    documents=splits,
    embedding=embedding,
    persist_directory=persist_directory
)

𝗦𝘁𝗲𝗽 𝟲 : 𝗙𝗶𝗻𝗱 𝘀𝗶𝗺𝗶𝗹𝗮𝗿 𝗿𝗲𝗹𝗮𝘁𝗲𝗱 𝗰𝗼𝗻𝘁𝗲𝗻𝘁 𝗳𝗿𝗼𝗺 𝘁𝗵𝗲 𝗱𝗯

question = "What would be a memorable valentine gift for a woman aged 39 years?"

docs = vectordb.similarity_search(question, k=2)
vectordb.persist()
content = docs[0].page_content

print(content)

I get :—

Valentine’s Day flowers
 
and a romantic homemade dinner.

So it seems from my knowledge base that flowers and a romantic homemade dinner might be a good idea.

𝗦𝘁𝗲𝗽 𝟳 : 𝗕𝘂𝗶𝗹𝗱 𝗮 𝗿𝗲𝘁𝗿𝗶𝗲𝘃𝗲𝗿 𝘁𝗼 𝗾𝘂𝗲𝗿𝘆 𝘁𝗵𝗲 𝗟𝗟𝗠 𝗮𝗻𝗱 𝘀𝘂𝗽𝗽𝗼𝗿𝘁 𝘁𝗵𝗲 𝗺𝗼𝘀𝘁 𝗿𝗲𝗹𝗲𝘃𝗮𝗻𝘁 𝗿𝗲𝘀𝘂𝗹𝘁 𝗳𝗿𝗼𝗺 𝘁𝗵𝗲 𝗩𝗲𝗰𝘁𝗼𝗿 𝗗𝗕

# build some metadata to support the query to the LLM
# the preferable source file of the gift
metadata_field_info = [
    AttributeInfo(
        name="source",
        description="The source of the gift. One of ['file3.pdf']",
        type="string",
    ),
] 

document_content_description = "memorable gift for a woman"
llm = OpenAI(temperature=0)
retriever = SelfQueryRetriever.from_llm(
    llm,
    vectordb,
    document_content_description,
    metadata_field_info,
    verbose=True
)

#get the most relevant result for the query based on our own corpus
question = "What would be a memorable valentine gift for a woman aged 39 years?"
docs = retriever.get_relevant_documents(question)

print(docs)

I get :—

page_content='Valentine’s Day flowers\n \nand a romantic homemade dinner.'

So it seems (according to the knowledge base) that a combination of flowers and a home made dinner would be an excellent choice.

𝗦𝘁𝗲𝗽 𝟴 : 𝗟𝗲𝘁𝘀 𝗯𝘂𝗶𝗹𝗱 𝘁𝗵𝗲 𝗽𝗿𝗼𝗺𝗽𝘁

template = """Given a person with the following profile:
{user_profile}
Use the following pieces of context to answer the question at the end.
{context}
Question: {question}
Helpful Answer:"""

user_profile = {
    "age": 39,
    "gender": "female",
    "interests": ["reading", "gardening", "music", "clothes", "outdoors"],
    "profession": ["teacher"]
}

# add context to the query template
prompt = PromptTemplate(template=template, input_variables=["user_profile", "context", "question"])

𝗦𝘁𝗲𝗽 𝟵 : 𝗟𝗲𝘁𝘀 𝗳𝗶𝗻𝗮𝗹𝗹𝘆 𝗮𝘀𝗸 𝗖𝗵𝗮𝘁𝗚𝗣𝗧

llm = ChatOpenAI(model_name="gpt-3.5-turbo-0125", temperature=0.0002)
llm_chain = LLMChain(prompt=prompt, llm=llm)

question = "What would be a memorable valentine's day gift for my wife aged 39?"

generated = llm_chain.run(user_profile=user_profile, context = docs, question = question)

# generate results
print("Gift ideas: " + generated)

And I get :—

Gift ideas: Based on the profile of your wife, a memorable Valentine's Day gift could be a combination of things she enjoys such as a book from her favorite author, a new plant or gardening tool for her garden, a vinyl record of her favorite music artist, a stylish piece of clothing, or a gift card for outdoor activities or experiences. You could also consider planning a romantic homemade dinner or surprising her with Valentine's Day flowers. Ultimately, the most memorable gift will be something thoughtful and personalized to her interests and preferences.

I actually followed the advice.

What do you think?

Which gift did I get her?

And Did she like the gift?

Follow me Ritesh Shergill

for more articles on

👨‍💻 Tech

👩‍🎓 Career advice

📲 User Experience

🏆 Leadership

I also do

✅ Career Guidance counselling — https://topmate.io/ritesh_shergill/149890

✅ Mentor Startups as a Fractional CTO — https://topmate.io/ritesh_shergill/193786