Skip to main content

​
Overview

RAG Chat allows you to upload PDF documents that will be processed, chunked, and stored in a vector database for intelligent question answering. This guide walks you through the upload process and best practices.

​
Upload Interface

The file upload interface is located in the sidebar of the RAG Chat application:
app.py
with st.sidebar:
    st.header('Upload de arquivos')
    uploaded_files = st.file_uploader(
        label='Faça aqui o upload dos seus arquivos: ',
        accept_multiple_files=True,
        type='pdf',
    )
You can upload multiple PDF files at once. All files will be processed and added to the same vector store.

​
How Document Processing Works

1

File Upload

Select one or more PDF files using the file uploader in the sidebar.
2

PDF Loading

Each file is processed using PyPDFLoader, which extracts text from all pages:
app.py
def process_file(file):
    with NamedTemporaryFile(delete=False, suffix='.pdf') as temp_file:
        temp_file.write(file.read())
        temp_file_path = temp_file.name
        loader = PyPDFLoader(temp_file_path)
        docs = loader.load()
3

Text Chunking

The extracted text is split into manageable chunks for better retrieval:
app.py
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size = 1000,
    chunk_overlap = 400
)
chunks = text_splitter.split_documents(docs)
  • chunk_size: 1000 characters per chunk
  • chunk_overlap: 400 characters overlap between chunks to preserve context
4

Vector Store Creation

Chunks are embedded and stored in a Chroma vector database:
app.py
def add_to_vector_store(documents, vector_store = None):
    if vector_store:
        vector_store.add_documents(documents)
    else:
        vector_store = Chroma.from_documents(
            documents=documents,
            embedding=OpenAIEmbeddings(),
            persist_directory=persistant_directory
        )
    return vector_store

​
Processing Flow

When you upload files, the system processes them automatically:
app.py
with st.spinner('Carregando arquivos...'):
    all_chunks = []
    for uploaded_file in uploaded_files:
        chunks = process_file(uploaded_file)
        all_chunks.extend(chunks)

    if all_chunks:
        vector_store = add_to_vector_store(
            vector_store = vector_store,
            documents = all_chunks
        )

​
Best Practices

Optimal Document Types:
  • Technical documentation
  • Research papers
  • Reports and white papers
  • Books and manuals
  • Any text-heavy PDF content
Avoid:
  • Image-only PDFs (without OCR)
  • Heavily formatted documents with complex layouts
  • Password-protected PDFs
  • Scanned documents without text layer

​
Tips for Best Results

  1. Multiple Related Documents: Upload documents on related topics together for better context understanding
  2. Clean PDFs: Use PDFs with clear text extraction (not scanned images)
  3. Reasonable Size: While there’s no strict limit, extremely large documents may take longer to process
  4. Incremental Uploads: You can upload additional documents at any time - they’ll be added to the existing vector store

​
Persistent Storage

Documents are stored persistently in the db directory:
app.py
persistant_directory = 'db'

def load_existing_vector_store():
    if os.path.exists(persistant_directory):
        vector_store = Chroma(
            persist_directory=persistant_directory,
            embedding_function=OpenAIEmbeddings()
        )
        return vector_store
    return None
Once uploaded, documents persist across sessions. You don’t need to re-upload them every time you restart the application.

​
Troubleshooting

​
Upload Not Working

  • Ensure the file is a valid PDF
  • Check that the PDF is not password-protected
  • Verify you have sufficient disk space

​
Slow Processing

  • Large PDFs take longer to process
  • The spinner shows “Carregando arquivos…” during processing
  • Wait for the process to complete before asking questions

​
Next Steps

Asking Questions

Learn how to ask questions about your uploaded documents

Configuration

Customize chunk size and other processing parameters

Build docs developers (and LLMs) love

Get started for freeTalk to us