The Energy Wizard

This example demonstrates how to download a set of PDFs, convert to text, chunk, embed, and then run a streamlit app that interfaces an LLM with the text corpus.

Notes:

• In this example, we use the optional popper PDF utility which you will have to install separately. You can also use the python-native PyPDF2 package when calling using elm.pdf.PDFtoTXT but we have found that poppler works better.

• Streamlit is required to run this app, which is not an explicit requirement of this repo (pip install streamlit)

• You need to set up your own OpenAI or Azure-OpenAI API keys to run the scripts.

Downloading and Embedding PDFs

Run python ./retrieve_docs.py to retrieve 20 of the latest NREL technical reports from OSTI. The script then converts the PDFs to text and then runs the text through the OpenAI embedding model.

Running the Streamlit App

Run streamlit run ./run_app.py to start the streamlit app. You can now chat with the Energy Wizard, which will interface with the downloaded text corpus to answer your questions.