Iraqi Legal GPT

This paper proposes and develops a small language model artificial intelligence (AI) system or chatbots capable of providing competent legal information to users within the Iraqi legal jurisprudence. One of the major contributions of the papers is addressing critical challenges related to the current Large Language Model (LLM) like ChatGPT of their enormous size and lack of localized knowledge of internal processes and lifestyles. The proposed model provides low latency time with limited economic resources and the availability of efficient legal professionals for understanding the legal and rights requirements of individuals within the Iraqi region which can be run on a local computer without an internet connection. The research leverages the open source h2ogpt as a backbone language model and was trained with aggregated and curated legal documents from the Iraqi legal system. This approach allows users to input questions, and through the aid of Embedding algorithms and Vector Store algorithms, the data is stored and converted into a machine-readable format for subsequent comprehension and processing by the computer. The findings indicate a promising foundation for the project, featuring a user-friendly interface (UI) and an updateable back-end to accommodate future enhancements. The system achieved a noteworthy 1-minute response time with an accuracy rate of 80%. Notably, the development objectives aimed to establish a locally deployable system that is easily accessible and modifiable to address evolving challenges in the current computational era of information dissemination.