A Knowledge-Retrieval Chatbot for Petroleum Engineering Using LLAMA and LangChain

The increasing use of Large Language Models (LLMs) in question-answering systems has raised concerns regarding response reliability, hallucinated content, and insufficient grounding in authoritative knowledge sources, particularly in specialized engineering domains such as petroleum engineering. In this study, a domain-specific petroleum engineering chatbot based on the Retrieval-Augmented Generation (RAG) framework on some petroleum books and drilling industry standards is developed to provide accurate and reliable responses grounded in industrial standards. The system employs LLAMA ۳.۲ with ۷ billion parameters, deployed locally on a private server, and utilizes a curated collection of petroleum engineering standards as its knowledge base. Text segmentation is performed using RecursiveTextSplitter, and key parameters related to chunking and model inference are empirically optimized to improve retrieval accuracy and response coherence. The LangChain framework is used to integrate the retrieval and generation components, while Streamlit provides an interactive user interface. System performance is evaluated using a benchmark dataset of domain-specific question-answer pairs, along with an analysis of user interactions and satisfaction. The results indicate that the proposed approach delivers more coherent, accurate, and trustworthy responses compared to general-purpose language models. Overall, the study demonstrates that combining open-source LLMS with RAG and local deployment offers an effective and scalable solution for intelligent decision-support systems in the oil and gas industry.

کلیدواژه ها:

Large Language Models ، Retrieval-Augmented Generation ، Knowledge-Based Chatbot ، Langchain