app.py

from langchain_groq import ChatGroq
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain_community.document_loaders import TextLoader, PyPDFLoader
from langchain.text_splitter import CharacterTextSplitter
from langchain.chains import RetrievalQA
import os

# Clave de API de OpenAI
from groq import Groq


from dotenv import load_dotenv
_ = load_dotenv()

# 1. Carga ambos documentos - CV y descripción del trabajo
# Job Description
jd_loader = TextLoader("JD/01_falabella_product_owner.txt")
jd_documents = jd_loader.load()

from langchain_community.document_loaders import PyPDFLoader

# Resume (PDF)
resume_loader = PyPDFLoader("docs/franco_resume_en.pdf")
resume_documents = resume_loader.load()

# Combina ambos documentos
all_documents = jd_documents + resume_documents

# Model context: 8,192 tokens
# (roughly 6,000-8,000 characters
# depending on language)

# Good chunk size options for your model:

# 2000-3000 characters: Good balance of context and efficiency
# 4000 characters: Maximum recommended for your model

# Total context usage: chunks + question + system prompts

# Keep overlap at 10-20% of chunk size for better continuity

chunk_size = 2500
chunk_overlap = int(chunk_size * 0.15)
text_splitter = CharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap)
docs = text_splitter.split_documents(all_documents)

# 2. Crea los embeddings y el vectorstore
embedding_model = HuggingFaceEmbeddings(
    model_name="sentence-transformers/all-MiniLM-L6-v2"
)

# Alternative embedding models you can use:

# all-MiniLM-L6-v2 (recommended - small and fast)
# all-mpnet-base-v2 (better quality, larger)
# paraphrase-multilingual-MiniLM-L12-v2 (if you need multilingual support)

db = FAISS.from_documents(docs, embedding_model)

# 3. Configurar el modelo LLM de Groq
llm = ChatGroq(
    model="llama3-70b-8192",
    temperature=0,
    max_tokens=None,
    timeout=None,
    max_retries=2,
    # other params...
)

# 4. Configura el modelo y el sistema de recuperación
retriever = db.as_retriever()
qa_chain = RetrievalQA.from_chain_type(llm=llm, retriever=retriever)

# 5. Haz una pregunta
#query = "¿Cuál es el tema principal del documento?"
#response = qa_chain.invoke({"query": query})

#print("Respuesta:", response)

#query = "What are Franco's skills?"
#response = qa_chain.invoke({"query": query})

#print("Respuesta:", response)

query = "¿puede Franco Cedillo postular al puesto de Product Owner en Falabella?"
response = qa_chain.invoke({"query": query})

query =response['query']
print("query:", query)

result =response['result']
print("result:", result)