This preliminary study project aims to enhance the reliability of GIT SA’s accounting document scans by combining artificial intelligence and human expertise. Currently, many automated tools make errors when analyzing various documents, which can slow down the work of SMEs and accounting firms. The objective is to create a more precise, secure, and adaptable solution for local needs.
To achieve this, the project is developing a system called Human Augmented RAG (HRAG). It uses advanced artificial intelligence models to read and understand documents, but also incorporates user corrections and feedback to continuously improve the accuracy of the results. This method allows adaptation to different document types without needing to fully retrain the models and ensures that all data remains confidential, without resorting to closed online solutions.
The preliminary study focuses on two key areas:
- AI Model Testing:
Analysis of the performance of various advanced models and contextual learning techniques on a corpus of anonymized real documents. The goal is to identify the most effective combination to improve automatic recognition. - Technical Study and Design:
Design of the HRAG system architecture, defining the necessary resources and estimating development efforts. The team also proposes a roadmap for future deployment, with recommendations for a large-scale Innosuisse project.
This exploratory phase has enabled GIT SA to make its Peppermint software more reliable and competitive, while laying the groundwork for an intelligent solution tailored to the real needs of users.
