IntelliAnswer

Goal:

A system that processes user-provided question files and supplementary documents. It extracts questions, answers them using information from the supplementary files when available, and falls back to an LLM for answers when necessary.

Implemented Features:

• Read docx files.

  If you want to use the pdf file then instead of using the read_docx function, you can use the bellow function

  from langchain_community.document_loaders import UnstructuredPDFLoader
  def load_pdf(file_path):
    loader = UnstructuredPDFLoader(file_path=file_path)
    documents = loader.load()
    print(f"Loaded {len(documents)} documents")
    return documents

• Extract information based on user query (currently assessment's task 1 questions )

  def extract_questions(qa_chain):
    # change the query according to your task
    query = """
    [INST] Based on the content of the document, find all the questions for assesment task 1.
    Format your response as a numbered list. [/INST]
    """
    result = qa_chain({"query": query})
    return result["result"]

Installation

1. Create a virtual environment (optional but recommended)

   python -m venv llmrag

2. Install all the dependencies

   pip install -r requirements.txt

3. Download Ollama from here [https://ollama.com/download]

4. Run ollama after installing

5. In terminal you need to pull the llama and nomic-embed-text. Although you can use any of the model available in the ollama repository.

   ollama run llama3
   ollama pull nomic-embed-text

6. Verify your installation

   ollama list

7. Now run the python file. For instance, you can use the following command to run the langchain_ollama_llama3_rag_for_docx.py script.

   python3 langchain_ollama_llama3_rag_for_docx.py

Note:

• Before running the script, you must specify the filepath in the main function.

• If your docx file is large enough, then try to tweak the chunk_size and chunk_overlap parameters accordingly.

  def split_documents(documents):
      text_splitter = RecursiveCharacterTextSplitter(chunk_size=2000, chunk_overlap=1000)
      chunks = text_splitter.split_documents(documents)
      document = chunks[0]
      print(document.page_content)
      print(document.metadata)
      print(f"Split into {len(chunks)} chunks")
      return chunks