Skip to content

Latest commit

 

History

History
120 lines (88 loc) · 3.35 KB

README.md

File metadata and controls

120 lines (88 loc) · 3.35 KB

Neural Wave Hackaton

  • Team Name: <teamName>
  • Project Name: Swisscom project <teamName>
  • Hackathon: NeuralWave
  • Date: October 25-27, 2024
  • Team Members:
    • Tornike Onoprishvili
    • Riccardo Sacco
    • Carla Lopez Zurita
    • Roberts Kalvitis
    • Michele Smaldone

Project Description

While Swisscom has a fully functional chatbot implemented on their website, we believe there is still room for improvement. The original chatbot sometimes presents issues with consistency when the user switches languages, which is a common occurrence given Switzerland's multilingual setting. Other considerations include providing helpful references within the Swisscom website, as well as offering truly useful responses to users' questions. Our goal is to create an effective chatbot that assists users with a variety of issues, beating performance of the currently implementing chatbot. The chatbot should:

  • Detect language and apply it to conversation correctly;
  • Filter out irrelevant user questions;
  • Give users useful responses that are based on Swisscom’s publicly available data on their website.
Workflow
Code Workflow

Project report

Can be found at ./Project Report.pdf.

Project video

Can be found at link.

Project frontend

https://8001-01jb2c5awq5kbbyy3p0g1jmccn.cloudspaces.litng.ai

Folder structure

    ├── assets
    │   └── prompts.md
    ├── chroma
    ├── data
    │   ├── botresults
    │   │   └── results_on_tests.json
    │   ├── inputs.json
    │   ├── output.json
    │   └── test_data.json
    ├── notebooks
    │   └── evaluate_results.ipynb
    ├── src
    │   ├── __init__.py
    │   ├── ai_judge.py
    │   ├── chat.py
    │   ├── evaluator.py
    │   ├── ingest_documents.py
    │   ├── make_predictions.py
    │   ├── swisscom_rag_chat.py
    │   └── swisscom_rag.py
    ├── LICENSE.md
    ├── README.md
    ├── Project Report.pdf
    └── requirements.txt

Installing Dependencies

To install the necessary packages listed in requirements.txt, run the following command:

pip install -r requirements.txt

Create a .env File

  1. In the root directory of your project, create a new file named .env. Use .env.example as an example.
  2. Open the .env file and add your API key as follows:
OPENAI_API_KEY=your_api_key_here

Download the dataset

  1. Download the dataset from https://swisscom-my.sharepoint.com/:u:/p/stefano_taillefert/EegWIyF8835PuUXsyuzmGGsBcxu7gFVcJVhyOpLVhZ_g4A?e=nsivZN
  2. place it in the root directory and run command in terminal.
unzip dataset.zip

Running json evaluation

  1. Place the input.json in data/
  2. Run predictions
python src/make_predictions.py
  1. Results are in data/output.json

Running project chat from terminal

Ingest the document base:

python src/ingest_documents.py

Run chat:

python src/chat.py