This repository contains a collection of solutions to the first round of enterprise-rag-challenge.
In a nutshell, the enterprise-rag-challenge is about building a piece of software that answers questions like
How much did the company "Accuray Incorporated" spend on risk management in 2022?
accurately. The data containing the basic factual information is provided in unstructured PDFs. As such, the expectation is that solutions will use large language models (LLMs) in some form to solve the challenge - hence the term RAG (Retrieval-Augmented Generation) in its name.
The solutions in this repository were developed quite quickly - within 2 work days - and should therefore be viewed as proof-of-concept rather than ready-for-production. And as expected, each solution used LLMs in some way, but it turned out that how the LLMs are used differ quite a bit across the solutions.
Thus, we expect this repository to be most interesting to developers looking for inspiration on possible approaches to Retrieval-Augmented Generation.
Each directory in this repository represents a solution. Each solution is self-contained; refer to the included READMEs for more information.
- Daniel - simple solution using checklist pattern with structured outputs. First place.
- Felix - multi-agent solution with ChatGPT-4o. 12th place.
- Maria - solution using OpenAI Assistants API. 13th place.
- Pedro - locally-capable solution using openchat-3.5-0106. ninth place.
Happy Browsing!
We would like to thank our employer Trustbit (now a part of TIMETOACT GROUP as TimeToAct Austria) for providing us with time & resources to work on this challenge.