AIM Hackathon Oct 2024 - Template

Repository for the first AIM Hackathon together with TIMETOACT GROUP Österreich on 19.10.2024

Check out OpenAI pricing

https://openai.com/api/pricing/

Set up OpenAI API Key

Copy your teams API key from the slack channel description and place it in the .env_template file.

Don't forget to replace the filename to .env afterwards!

Check out the sample code to see how to load the key.

About the data

The dataset reports.json contains the following keys:

• company_name: name of the company
• year: year of the report (if report is for two years, the first year is used)
• dataset: subdataset name
  • scraped: scraped from a website, might contain broken links and irrelevant pdfs (90 samples, 2016-2024)
  • austria: reports from mainly Austrian companies (33 samples)
  • handcrafted: selected by hand, ESG and sustainability reports (23 samples, 2019, 2021, 2023 for every company)
• pdf_url: link to the pdf report

Jump start

Fork this repository

Simply fork this repository to start working on your project.

Set up environment

Create a new environment (e.g. with conda)

conda create -n aim_hackathon_oct24 python=3.10

Install the requirements

pip install -r requirements.txt

Sample code

There is a super simple RAG implementation to help getting you started in sample_code.ipynb.

Hints

Very simple RAG pipeline to start with.

You can extract openAI API token usage from the response with response['usage'].

You can use tiktoken to manually count tokens of a string:

import tiktoken
tokenizer = tiktoken.get_encoding("o200k_base")  # for gpt 4o

Structured outputs force the LLM to output e.g. only integers.