LANGCHAIN — Integrating ChatGPT with Google Drive and Notion Data

Computer science is no more about computers than astronomy is about telescopes. — Edsger W. Dijkstra.

Integrating ChatGPT with Google Drive and Notion Data

ChatGPT has become an essential tool in various workflows due to its versatile nature and general-purpose performance. However, to achieve the best results, users often need to provide context and instructions through prompting. This prompting process can be laborious, involving manual extraction of relevant information from multiple documents to supply the necessary context to ChatGPT.

To address this challenge, the Noah personal assistant app was developed in collaboration with the Tavrn team. Noah leverages LangChain and LangSmith technologies to efficiently retrieve context from Google Drive and Notion, enabling users to experience an AI copilot that seamlessly provides relevant context to answer queries.

Below, we’ll explore how to integrate ChatGPT with Google Drive and Notion data using LangChain’s Noah.

Syncing Files from Google Drive and Notion

To get started with Noah, users can sync files from their Google Drive and Notion accounts. The process involves selecting the desired tools and choosing specific files or quickly selecting a set number of recent files from Google Drive or Notion.

# Code snippet to sync files from Google Drive or Notion
def sync_files(tool, file_selection):
    if tool == "Google Drive":
        # Code to sync files from Google Drive
        pass
    elif tool == "Notion":
        # Code to sync files from Notion
        pass

Processing Documents

Once the files are selected, Noah processes the documents using optimized, context-aware document loaders and state-of-the-art embeddings models. This involves semantic chunking and utilizing LangChain’s CharacterTextSplitter for efficient processing of various document types, such as spreadsheets, documents, PDFs, and slides.

# Code snippet for processing documents
def process_documents(files):
    for file in files:
        # Code to process each document using optimized loaders and embeddings models
        pass

Retrieving Relevant Content

When a user asks a question, Noah fetches the most relevant content across multiple sources using cosine similarity vector search and passes them to multi-chain LLM calls to obtain the best possible answer. Additionally, Langsmith is utilized for fine-tuning which chains and prompts to use for the final user answer, ensuring optimal memory parameter for ConversationBufferWindowMemory for reliable answers.

# Code snippet for retrieving relevant content and obtaining the answer
def retrieve_and_answer(question):
    relevant_content = fetch_relevant_content(question)
    answer = obtain_best_answer(relevant_content)
    return answer

Filtering and Providing Answer

After retrieving the relevant content, Noah passes the chunks into an intermediary, GPT-4 powered chain to filter out any conflicting information, prioritizing more recent sources. Finally, Noah provides the answer, with the appropriate sources cited.

# Code snippet for filtering and providing the answer
def filter_and_provide_answer(chunks):
    filtered_chunks = filter_conflicting_information(chunks)
    final_answer = generate_answer(filtered_chunks)
    return final_answer

By integrating ChatGPT with Google Drive and Notion data using Noah, users can streamline the context fetching process and leverage the power of LangChain to boost their productivity in various workflows.