Documentation for AI Chatbot "DOC chat" with PDF Knowledge Base

Overview

This application is designed to interact with users and answer questions based on the content of uploaded PDF documents. The chatbot is equipped with fallback logic to handle out-of-context questions gracefully. The application leverages Groq AI inference technology, FAISS for similarity search, and a user-friendly interface created with Streamlit.

---

Key Features

1. PDF Knowledge Base:

   • Users can upload multiple PDFs to serve as the chatbot's knowledge base.
   • The application processes the PDFs to extract content and create a vector database for efficient querying.

2. Intelligent Question Answering:

   • Answers user queries based on the PDF content.
   • Utilizes Groq AI models for generating responses.

3. Fallback Handling:

   • When a question is out of context or unrelated to the PDF content, the chatbot responds: “Sorry, I didn’t understand your question. Do you want to connect with a live agent?”

4. Interactive Interface:

   • Provides a seamless, text-based chat interface.
   • Allows users to upload PDFs, ask questions, and view answers.

5. Reference Context:

   • Displays the source of information for each response when available.

---

Architecture and Workflow

1. PDF Upload and Processing

• Step: Users upload one or more PDFs via the interface.
• Action: The uploaded PDFs are processed using FAISS to create a vector database for fast similarity-based search.

2. Chat Interface

• Step: Users ask questions through the chat input.
• Action:
  • The question is processed against the vector database.
  • Relevant context is retrieved and passed to the LLM via a prompt template.
  • The LLM generates a response based on the context.

3. Fallback Mechanism

• Step: If no relevant context is retrieved or the question is unrelated, the chatbot generates a fallback response.

4. Reference Display

• Step: Users can view the references for responses.
• Action: Contextual excerpts from the PDFs are displayed for transparency.

---

Tools and Libraries Used

1. Streamlit:

   • Used for building the user interface.
   • Provides features like file upload, chat interface, and navigation menus.

2. Groq AI:

   • Powers the chatbot's inference capabilities.
   • Supports various advanced LLMs such as Llama3-8b-8192 and Mixtral-8x7b-32768.

3. FAISS:

   • Handles vector storage and similarity search for efficient information retrieval.

4. LangChain:

   • Utilized for prompt management and integration with Groq AI.

5. Python:

   • Core programming language used to implement the solution.

---

Implementation Details

1. File Upload and Processing

• Code:

st.subheader("Upload PDF(s)")
pdf_docs = st.file_uploader("Upload your PDFs", type=['pdf'], accept_multiple_files=True)

if st.button("Process", type="primary", disabled=not pdf_docs):
    with st.spinner("Processing..."):
        try:
            st.session_state.vector_store = create_vectorstore(pdf_docs)
            st.session_state.prompt = True
            st.success("Database is ready!")
        except Exception as e:
            st.error(f"Failed to process PDFs: {e}")

• Explanation:
  • PDFs are uploaded and processed into a vector database using the create_vectorstore function.
  • The application indicates success or failure during processing.

---

2. Chat Functionality

• Code:

question = st.chat_input(
    placeholder="Enter your question related to uploaded document",
    disabled=not st.session_state.prompt,
)

if question:
    st.session_state.messages.append({"role": "user", "content": question})
    st.chat_message("user").write(question)

    with st.spinner("Processing..."):
        try:
            response = get_llm_response(llm, prompt_template, question)
            answer = response.get("answer", "No answer provided.")
            st.session_state.messages.append({"role": "assistant", "content": answer})
            st.chat_message("assistant").write(answer)
        except Exception as e:
            st.error(f"Failed to get response: {e}")

• Explanation:
  • User input is processed to retrieve context and generate an answer.
  • Responses are displayed in the chat interface.

---

3. Fallback Logic

• Code:

prompt_template = ChatPromptTemplate.from_template(
    """
    Answer the question based on the provided context only. If question is not within the context, do not try to answer
    and respond that "Sorry, I didn’t understand your question. Do you want to connect with a live agent?".
    Please provide the most accurate response based on the question.
    <context>
    {context}
    Questions: {input}
    """
)

• Explanation:
  • The prompt ensures the chatbot only responds if the question is within the provided context.
  • If not, a fallback response is generated.

---

4. References

• Code:

if st.session_state.response and "context" in st.session_state.response:
    for i, doc in enumerate(st.session_state.response["context"]):
        with st.expander(f"Reference #{i + 1}"):
            st.write(doc.page_content)
else:
    st.info("No references available. Upload documents and ask a question first.")

• Explanation:
  • References are extracted from the context and displayed in an expandable section.

---

Challenges and Solutions

1. Handling Large PDFs:

   • Challenge: Processing large documents efficiently.
   • Solution: FAISS was used for similarity-based querying to speed up content retrieval.

2. Ensuring Accurate Responses:

   • Challenge: Preventing irrelevant answers.
   • Solution: Context-bound prompts were used to limit the chatbot's responses.

3. User-Friendly Interface:

   • Challenge: Designing an intuitive interface.
   • Solution: Streamlit was used to create a clear, interactive UI.

---

Instructions to Run the Application

1. Install dependencies:

   pip install streamlit langchain groq-python faiss-cpu python-dotenv

2. Set up environment variables in a .env file:

   GROQ_API_KEY=your_api_key

3. Set up environment variables in a .env file:

   INFERENCE_API_KEY=your_api_key

4. Run the application:

   streamlit run app.py

5. Upload PDFs and start asking questions!

---

Future Improvements

1. Add support for other document formats (e.g., Word, Excel).
2. Enhance accuracy by fine-tuning the prompt or model.
3. Implement a live agent connection for unresolved queries.