Chat with PDF using Pinecone

SUTRA

SUTRA by TWO Platforms

SUTRA is a family of large multi-lingual language models (LMLMs) pioneered by Two Platforms. SUTRA’s dual-transformer approach extends the power of both MoE and Dense AI language model architectures, delivering cost-efficient multilingual capabilities for over 50+ languages. It powers scalable AI applications for conversation, search, and advanced reasoning, ensuring high-performance across diverse languages, domains and applications.

Get Your API Keys

Before you begin, make sure you have:

  1. A SUTRA API key (Get yours at TWO AI's SUTRA API page)
  2. Basic familiarity with Python and Jupyter notebooks

This notebook is designed to run in Google Colab, so no local Python installation is required.

📌 1. Install Required Packages

!pip install -q langchain langchain_openai langchain-community requests pypdf langchain-pinecone ipywidgets

📌 STEP 2 : Setup API Keys

import os
from google.colab import userdata

# Set the API key from Colab secrets
os.environ["SUTRA_API_KEY"] = userdata.get("SUTRA_API_KEY")
os.environ["OPENAI_API_KEY"] = userdata.get("OPENAI_API_KEY")
os.environ["PINECONE_API_KEY"] = userdata.get("PINECONE_API_KEY")

📌 STEP 3 : Load Your PDF Document

from langchain_community.document_loaders import PyPDFLoader

# Load PDF using PyPDFLoader
loader = PyPDFLoader("/content/NIPS-2017-attention-is-all-you-need-Paper.pdf")  # Replace with your actual PDF path
documents = loader.load()
print(f"Loaded {len(documents)} pages.")

📌 STEP 4 : Split Document into Chunks

from langchain.text_splitter import RecursiveCharacterTextSplitter

# Split documents into chunks
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=1000,
    chunk_overlap=100
)
chunks = text_splitter.split_documents(documents)
print(f"Split into {len(chunks)} chunks.")

📌 STEP 5 : Set Up Pinecone Vector Database

import os
from pinecone import Pinecone, ServerlessSpec

pc = Pinecone(api_key=os.getenv("PINECONE_API_KEY"))
index_name = "sutra-pdf-indexs1"

# Create index if it doesn't exist
if index_name not in pc.list_indexes().names():
    pc.create_index(
        name=index_name,
        dimension=1536,
        metric="cosine",
        spec=ServerlessSpec(cloud="aws", region="us-east-1")
    )
index = pc.Index(index_name)

📌 STEP 6 : Set Up Pinecone Vector Store

from langchain_openai import OpenAIEmbeddings
from langchain_pinecone import PineconeVectorStore

# Create vector store using Pinecone
embeddings = OpenAIEmbeddings(api_key=os.getenv("OPENAI_API_KEY"))
vectorstore = PineconeVectorStore(index, embeddings)

# Add documents to Pinecone vector store
vectorstore.add_documents(chunks)

📌 STEP 7 : Set Up Conversation Memory and RAG Chain

from langchain.memory import ConversationBufferMemory
from langchain.chains import ConversationalRetrievalChain
from langchain_openai import ChatOpenAI

# Set up conversation memory
memory = ConversationBufferMemory(
    memory_key="chat_history",
    return_messages=True
)

# RAG Chain with Sutra LLM
rag_chain = ConversationalRetrievalChain.from_llm(
    llm=ChatOpenAI(
        api_key=os.getenv("SUTRA_API_KEY"),
        base_url="https://api.two.ai/v2",
        model="sutra-v2",
        temperature=0.7
    ),
    retriever=vectorstore.as_retriever(),
    memory=memory
)

📌 STEP 8 : Ask Question with Language Specification

from langchain.schema import HumanMessage

# Desired language for the response
language = "Hindi"  # change to any supported language

# User question
question = "What is Transformer ?"

# Get RAG answer from chain
context_result = rag_chain.invoke({"question": question})
rag_context = context_result['answer']

# Format prompt for multilingual Sutra response
system_prompt = f"""
You are a helpful assistant answering based on a document.
Use this context: {rag_context}
Always reply in: {language}
Question: {question}
"""

# Invoke Sutra LLM directly for language-controlled response
llm = ChatOpenAI(
    api_key=os.getenv("SUTRA_API_KEY"),
    base_url="https://api.two.ai/v2",
    model="sutra-v2",
    temperature=0.5,
)

response = llm.invoke([HumanMessage(content=system_prompt)])
print("User Question:", question)
print(f"Assistant ({language}):", response.content.strip())

✅ Finally Integrated UI

# 1. Imports
import os
import ipywidgets as widgets
from IPython.display import display, HTML, clear_output
from tempfile import NamedTemporaryFile
from langchain_community.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
from langchain.memory import ConversationBufferMemory
from langchain.chains import ConversationalRetrievalChain
from langchain.schema import HumanMessage
from pinecone import Pinecone, ServerlessSpec
from langchain_pinecone import PineconeVectorStore


# 2. Get Sutra Chat Model
def get_sutra_model():
    return ChatOpenAI(
        api_key=os.getenv("SUTRA_API_KEY"),
        base_url="https://api.two.ai/v2",
        model="sutra-v2",
        temperature=0.7
    )

# 3. Load & Index PDF using Pinecone
def load_and_index_pdf(pdf_path):
    loader = PyPDFLoader(pdf_path)
    docs = loader.load()

    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
    chunks = text_splitter.split_documents(docs)

    # Embeddings
    embeddings = OpenAIEmbeddings(api_key=os.getenv("OPENAI_API_KEY"))

    # Pinecone setup
    pc = Pinecone(api_key=os.getenv("PINECONE_API_KEY"))
    index_name = "sutra-pdf-indexs2"

    # Create index if it doesn't exist
    if index_name not in pc.list_indexes().names():
        pc.create_index(
            name=index_name,
            dimension=1536,
            metric="cosine",
            spec=ServerlessSpec(cloud="aws", region="us-east-1")
        )

    index = pc.Index(index_name)
    vectorstore = PineconeVectorStore(index, embeddings)

    # Add to index
    vectorstore.add_documents(chunks)

    # RAG Chain
    memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
    chain = ConversationalRetrievalChain.from_llm(
        llm=get_sutra_model(),
        retriever=vectorstore.as_retriever(),
        memory=memory
    )
    return chain

# 4. UI Components
pdf_file_upload = widgets.FileUpload(
    accept='.pdf',
    multiple=False,
    description='📁 Upload PDF',
    layout=widgets.Layout(width='300px')
)

load_pdf_button = widgets.Button(
    description="🔄 Load PDF",
    button_style='info',
    layout=widgets.Layout(width='150px')
)

status_output = widgets.Output()

languages = [
    "English", "Hindi", "Gujarati", "Bengali", "Tamil", "Telugu", "Kannada",
    "Malayalam", "Punjabi", "Marathi", "Urdu", "Assamese", "Odia", "Sanskrit",
    "Korean", "Japanese", "Arabic", "French", "German", "Spanish", "Portuguese",
    "Russian", "Chinese", "Vietnamese", "Thai", "Indonesian", "Turkish", "Polish",
    "Ukrainian", "Dutch", "Italian", "Greek", "Hebrew", "Persian", "Swedish",
    "Norwegian", "Danish", "Finnish", "Czech", "Hungarian", "Romanian", "Bulgarian",
    "Croatian", "Serbian", "Slovak", "Slovenian", "Estonian", "Latvian", "Lithuanian",
    "Malay", "Tagalog", "Swahili"
]

lang_dropdown = widgets.Dropdown(
    options=languages,
    value="English",
    description='🌐 Language:',
    layout=widgets.Layout(width='300px')
)

chat_output = widgets.HTML(
    value="<div style='padding:10px; font-family:Arial; font-size:14px; height:300px; overflow-y:auto; border:1px solid #ccc; border-radius:5px;'>Chat history will appear here...</div>"
)

user_input = widgets.Text(
    placeholder='Type your message...',
    layout=widgets.Layout(flex='4', width='auto')
)

send_button = widgets.Button(
    description="📤 Send",
    button_style='primary',
    layout=widgets.Layout(flex='1', width='auto')
)

messages = []
conversation_chain = None

# 5. Load PDF Logic
def on_load_pdf(b):
    global conversation_chain
    uploaded_files = pdf_file_upload.value

    with status_output:
        clear_output()
        if not uploaded_files:
            print("❌ Please upload a PDF file first.")
            return
        try:
            print("⏳ Processing uploaded PDF...")
            uploaded_file = list(uploaded_files.values())[0]
            with NamedTemporaryFile(delete=False, suffix=".pdf") as tmp:
                tmp.write(uploaded_file['content'])
                tmp_path = tmp.name

            conversation_chain = load_and_index_pdf(tmp_path)
            print("✅ PDF loaded and indexed successfully!")
        except Exception as e:
            print("❌ Error:", e)

load_pdf_button.on_click(on_load_pdf)

# 6. Chat Interaction Logic
def on_send_click(b):
    global conversation_chain
    if conversation_chain is None:
        with status_output:
            clear_output()
            print("❌ Load a PDF first.")
        return

    user_text = user_input.value.strip()
    if not user_text:
        return

    lang = lang_dropdown.value
    messages.append(f"<b style='color:#13f22d;'>You:</b> {user_text}")

    context_response = conversation_chain.invoke(user_text)
    rag_context = context_response['answer']

    system_msg = f"""
You are a helpful assistant answering based on a document.
Use this context: {rag_context}
Always reply in: {lang}
Question: {user_text}
"""

    chat_model = get_sutra_model()
    sutra_response = chat_model.invoke([HumanMessage(content=system_msg)])
    assistant_reply = sutra_response.content.strip()

    messages.append(f"<b style='color:#007acc;'>Assistant ({lang}):</b> {assistant_reply}")
    chat_html = "<br>".join(messages)
    chat_output.value = f"<div style='padding:10px; font-family:Arial; font-size:14px; height:300px; overflow-y:auto; border:1px solid #ccc; border-radius:5px;'>{chat_html}</div>"
    user_input.value = ""

send_button.on_click(on_send_click)

# 7. Final UI Layout
input_row = widgets.HBox([user_input, send_button])
pdf_row = widgets.HBox([pdf_file_upload, load_pdf_button])

ui = widgets.VBox([
    widgets.HTML("<h3 style='font-family:Arial;'>📚 Multilingual Chat with PDF (Sutra + Pinecone)</h3>"),
    pdf_row,
    lang_dropdown,
    chat_output,
    input_row,
    status_output
])

# 8. Display App
display(ui)

On this page

SUTRA by TWO Platforms
Get Your API Keys
📌 1. Install Required Packages
📌 STEP 2 : Setup API Keys
📌 STEP 3 : Load Your PDF Document
📌 STEP 4 : Split Document into Chunks
📌 STEP 5 : Set Up Pinecone Vector Database
📌 STEP 6 : Set Up Pinecone Vector Store
📌 STEP 7 : Set Up Conversation Memory and RAG Chain
📌 STEP 8 : Ask Question with Language Specification
✅ Finally Integrated UI