import streamlit as st


def render_guide():

    st.write(
        """
When you send a query to LegisQA, it will attempt to retrieve relevant content from the past six congresses ([113th-118th](https://en.wikipedia.org/wiki/List_of_United_States_Congresses)) covering 2013 to the present, pass it to a [large language model (LLM)](https://en.wikipedia.org/wiki/Large_language_model), and generate a response. This technique is known as Retrieval Augmented Generation (RAG). You can read [an academic paper](https://proceedings.neurips.cc/paper/2020/hash/6b493230205f780e1bc26945df7481e5-Abstract.html) or [a high level summary](https://research.ibm.com/blog/retrieval-augmented-generation-RAG) to get more details. Once the response is generated, the retrieved content will be available for inspection with links to the bills and sponsors.


## Disclaimer

This is a research project. The RAG technique helps to ground the LLM response by providing context from a trusted source, but it does not guarantee a high quality response. We encourage you to play around, find questions that work and find questions that fail. There is a small monthly budget dedicated to the OpenAI endpoints. Once that is used up each month, queries will no longer work.


## Config

Use the `Generative Config` to change LLM parameters.
Use the `Retrieval Config` to change the number of chunks retrieved from our congress corpus and to apply various filters to the content before it is retrieved (e.g. filter to a specific set of congresses). Use the `Prompt Config` to try out different document formatting and prompting strategies.

    """
    )