Bureau Bot is a document chatbot built using the Streamlit framework, the LlamaIndex framework, and the OpenAI API. Its objective is to help users understand the documents from the Agentur für Arbeit by providing them with the information included in the official documents.
To run Bureau Bot, you need Python 3.10.12.
You can install the required packages using the command:
pip install -r requirements.txt
To use Bureau Bot run in the terminal:
chatbot_app.py
It will start a Streamlit web application that you can access in your browser at http://localhost:8501/.
To use Bureau Bot, you need to enter your OpenAi key.
Bureau Bot will display a chat of all the messages exchanged between the user and the chatbot.
In the Bureau Bot app, you can:
- Ask questions about the documents of the Agentur für Arbeit.
- Upload a document (in the formats .docx, .doc, or .pdf) by clicking "Choose a file from your hard drive"
Bureau Bot will return the pages and the documents used to create a response, you can download these documents by selecting the title of the file you want to download and clicking "Download".
To improve the chatbot, we will experiment with:
- Other language models: Llama2, Aleph Alpha, and other pre-trained models (e.g., Hugging Face).
- LangChain’s agents (e.g., web search retriever).
- More experiments with retrieval techniques (e.g., Karpathy’s SVM-based approach or Hypothetical Document Embedding).
- Finetune the embeddings with an open source LLM.
- Integrate Weights & Biases to monitor the LLM.
Bureau Bot was built by Daniele Didino and Marco Zausch as a portfolio project for the Data Science Retreat (Berlin). This project was mentored by Antonio Rueda-Toicen.