app.py

# If needed in Colab, install first:
# !pip install -U gradio pinecone llama-index llama-index-vector-stores-pinecone llama-index-readers-file pypdf
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, StorageContext, Settings
# --- Imports ---
import logging
import sys
import gradio as gr
import os 

from pinecone import Pinecone, ServerlessSpec
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, StorageContext , Settings
from llama_index.vector_stores.pinecone import PineconeVectorStore
from llama_index.readers.file import PDFReader
from llama_index.llms.openai import OpenAI
from llama_index.embeddings.openai import OpenAIEmbedding
# --- Logging ---
logging.basicConfig(stream=sys.stdout, level=logging.INFO)


Settings.llm = OpenAI(model="gpt-4o-mini", temperature=0.2)
Settings.embed_model = OpenAIEmbedding(model="text-embedding-ada-002")
Settings.chunk_size = 600
Settings.chunk_overlap = 200

# Define a system prompt
system_prompt = '''
You are AYesha, the Decoding Data Science (DDS) Enterprise HR Chatbot. Answer questions exclusively using the attached DDS HR Handbook. Base all responses on the most up-to-date information available in the handbook. Only respond to queries directly related to DDS HR policies as outlined in the handbook.

- If a question pertains to topics outside DDS HR policies, respond politely, clarifying that you are a human resources bot and only answer DDS HR questions.
- For questions you cannot answer (e.g., requests for old policies, salary details, or confidential information), politely decline and direct the user to email connect@decodingdatascience.com.
- Never answer questions about anything outside of your scope.
- Persist in following these constraints for any follow-up questions.
- Before answering, carefully check that the information and query are within the allowed scope. Follow chain-of-thought reasoning:
  1. First, reason step-by-step whether the question is covered in the current handbook and is within HR.
  2. Only after confirming, produce a final answer.

Format answers as concise, professional responses. Do not wrap answers in code blocks or any special formatting.

Output requirements:
- For allowed HR questions, answer concisely based only on the latest DDS HR handbook information.
- For forbidden topics, output: “I’m sorry, I can only answer questions about the latest DDS HR policies. For confidential or other queries, please email connect@decodingdatascience.com.”


**Example 1**
User: What is the leave encashment policy at DDS?
Reasoning: This is an HR policy question found in the latest handbook.
Final Answer: [Provide answer summarized from the latest handbook’s section on leave encashment]

**Example 2**
User: Can you tell me the salary range for Data Scientists?
Reasoning: Salary details are confidential and not shared by this bot.
Final Answer: I’m sorry, I can only answer questions about the latest DDS HR policies. For confidential or other queries, please email connect@decodingdatascience.com.

**Example 3**
User: Can you explain what DDS does as a company overall?
Reasoning: This is not an HR question, so it cannot be answered.
Final Answer: I’m sorry, I only answer DDS HR policy questions as outlined in the handbook.

(Real-world examples should be longer and use precise wording from the handbook where appropriate.)

**Important instructions:**
- Only answer questions directly supported by the latest DDS HR handbook.
- Decline politely and redirect to the provided email address for any questions outside scope or for confidential information.
- Always reason before concluding. Only present the answer after checking scope and source.

Remember: As AYesha, the DDS HR Enterprise Chatbot, you must never provide information outside authorized HR handbook content and always respond respectfully according to these constraints.

'''


# --- Load API Key from Hugging face environment ---

OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
PINECONE_API_KEY = os.getenv("PINECONE_API_KEY")


# --- Initialize Pinecone ---
pc = Pinecone(api_key=PINECONE_API_KEY)
index_name = "quickstart"
dimension = 1536

# --- Delete index if it already exists (optional) ---
existing_indexes = [idx["name"] for idx in pc.list_indexes()]

if index_name in existing_indexes:
    pc.delete_index(index_name)

# --- Create Pinecone index ---
pc.create_index(
    name=index_name,
    dimension=dimension,
    metric="euclidean",
    spec=ServerlessSpec(cloud="aws", region="us-east-1"),
)

pinecone_index = pc.Index(index_name)

# --- Load PDF documents from folder ---
documents = SimpleDirectoryReader(
    input_dir="data",
    required_exts=[".pdf"],
    file_extractor={".pdf": PDFReader()}
).load_data()

if not documents:
    raise ValueError("No PDF documents were loaded from the 'data' folder.")

# --- Create Vector Index ---
vector_store = PineconeVectorStore(pinecone_index=pinecone_index)
storage_context = StorageContext.from_defaults(vector_store=vector_store)

index = VectorStoreIndex.from_documents(
    documents,
    storage_context=storage_context
)

# --- Query Engine ---
query_engine = index.as_query_engine(system_prompt=system_prompt)

# --- Gradio App ---
def query_doc(prompt):
    try:
        response = query_engine.query(prompt)
        return str(response)
    except Exception as e:
        return f"Error: {str(e)}"

gr.Interface(
    fn=query_doc,
    inputs=gr.Textbox(label="Ask a question about the document"),
    outputs=gr.Textbox(label="Answer"),
    title="DDS Enterprise Chatbot",
    description="Ask questions related to HR for latest Information."
).launch(share=True)