Skip to main content

​
Overview

Once you’ve uploaded documents, you can ask questions about their content. RAG Chat uses your documents as context to provide accurate, grounded answers.

​
The Chat Interface

The main chat interface appears in the center of the application:
app.py
question = st.chat_input(placeholder='Digite aqui sua pergunta...')
You must upload at least one document before you can ask questions. The chat input only becomes active when documents are loaded.

​
How Question Answering Works

1

Ask Your Question

Type your question in the chat input box at the bottom of the screen.
2

Context Retrieval

The system searches the vector store for relevant document chunks:
app.py
def ask_question(model, query, vector_store):
    llm = ChatOpenAI(model = model)
    retriever = vector_store.as_retriever()
The retriever finds the most semantically similar chunks to your question.
3

Response Generation

The AI generates a response using the retrieved context:
app.py
chain = (
    {
        'context': retriever,
        'input': RunnablePassthrough()
    }
    | prompt
    | llm
)
response = chain.invoke(query)
4

Display Answer

The response is displayed in the chat interface with markdown formatting.

​
System Prompt

RAG Chat uses a specific system prompt to guide AI responses:
app.py
system_prompt = '''
Use o contexto para responder as perguntas.
Se não encontrar uma resposta no contexto,
explique que não há informações disponíveis.
Responda em formato de markdown e com visualizações
elaboradas e interativas.
Contexto: {context}
'''
The system is instructed to:
  • Only answer based on the provided context
  • Admit when information is not available
  • Format responses in markdown
  • Provide elaborate and interactive visualizations

​
Conversation History

RAG Chat maintains conversation history throughout your session:
app.py
if 'messages' not in st.session_state:
    st.session_state.messages = []

# Messages are preserved across interactions
messages = [('system', system_prompt)]
for message in st.session_state.messages:
    messages.append((message.get('role'), message.get('content')))
This allows for follow-up questions and contextual conversations.

​
Example Questions

​
Factual Questions

What are the main findings in this research paper?

How does the proposed algorithm work?

​
Comparative Questions

What are the differences between approach A and approach B?

Compare the results from chapters 3 and 5.

​
Analytical Questions

What are the limitations mentioned in the study?

Summarize the key takeaways from this document.

​
Follow-up Questions

Can you explain that in more detail?

What are some examples of this concept?

​
Understanding Responses

​
Markdown Formatting

Responses are rendered with full markdown support:
  • Bold and italic text
  • Bullet points and numbered lists
  • Code blocks
  • Tables
  • Headings

​
Grounded Answers

If the AI cannot find relevant information in your documents, it will explicitly state that the information is not available rather than making up an answer.

​
Context-Based Responses

All answers are based on the content of your uploaded documents. The AI:
  • ✅ Uses only information from your documents
  • ✅ Cites specific details from the context
  • ✅ Admits when information is unavailable
  • ❌ Does not use external knowledge
  • ❌ Does not make assumptions beyond the document content

​
Chat Workflow

app.py
if vector_store and question:
    # Display conversation history
    for message in st.session_state.messages:
        st.chat_message(message.get('role')).write(message.get('content'))

    # Display user question
    st.chat_message('user').write(question)
    st.session_state.messages.append({'role':'user', 'content': question})
    
    # Generate and display AI response
    with st.spinner('Buscando resposta...'):
        response = ask_question(
            model = selected_model,
            query = question,
            vector_store = vector_store
        )
        st.chat_message('ai').write(response)
        st.session_state.messages.append({'role':'ai', 'content':response})

​
Best Practices

​
Effective Question Strategies

Be Specific: Instead of “Tell me about the document,” ask “What methodology was used in the experiment?”
Use Context: Reference specific topics or sections: “What does the author say about machine learning in chapter 3?”
Ask Follow-ups: The conversation history allows you to build on previous answers

​
What to Expect

  1. Relevant Context: The retriever finds the most relevant chunks from your documents
  2. Synthesized Answers: The AI synthesizes information from multiple chunks if needed
  3. Honest Limitations: If information isn’t in your documents, the AI will tell you
  4. Rich Formatting: Responses use markdown for better readability

​
Troubleshooting

​
No Response Generated

  • Ensure documents are uploaded (check for “Carregando arquivos…” completion)
  • Verify your OpenAI API key is configured
  • Check that you’ve selected a model

​
Irrelevant Answers

  • Try rephrasing your question more specifically
  • Ensure your documents contain information about the topic
  • Consider uploading additional relevant documents

​
”Information Not Available” Responses

  • This means your question can’t be answered from the uploaded documents
  • Try uploading documents that contain the relevant information
  • Rephrase your question to align with document content

​
Next Steps

Model Selection

Choose the right GPT model for your needs

Configuration

Customize system prompts and other settings

Build docs developers (and LLMs) love

Get started for freeTalk to us