TLDR: Geolog-IA is an AI-powered conversational system developed by researchers at Universidad Central del Ecuador to provide natural language answers to questions about geology theses. It uses Llama 3.1 and Gemini 2.5 language models, combined with a Retrieval-Augmented Generation (RAG) architecture and an SQLite database, to overcome issues like hallucinations and outdated information. The system, evaluated with a BLEU score of 0.87, offers an intuitive web interface, making thesis information easily accessible for students, faculty, and administrators, and serves as a model for similar applications in other academic fields.
In today’s interconnected world, access to timely and relevant information is crucial, especially in academia. University theses are invaluable sources of knowledge, but often, accessing them can be a cumbersome process involving physical libraries, clunky digital repositories, or inefficient online search engines. These traditional methods often lead to wasted time and effort, making it difficult for students, faculty, and administrators to fully leverage the rich information contained within these documents.
Addressing this challenge, researchers at the Universidad Central del Ecuador have developed Geolog-IA, an innovative conversational system designed to simplify access to geology theses. This system acts like an intelligent assistant, allowing users to ask questions in natural language and receive accurate, contextualized answers about the university’s geology theses. You can find the full research paper here: GEOLOG-IA: SISTEMA CONVERSACIONAL SOBRE TESIS ACADÉMICAS.
How Geolog-IA Works
Geolog-IA is built on a sophisticated architecture that combines advanced artificial intelligence models with a structured database. At its core, it utilizes powerful language models like Llama 3.1 and Gemini 2.5. These models are enhanced by a technique called Retrieval-Augmented Generation (RAG), which allows the system to retrieve specific and up-to-date information from an external source—in this case, an SQLite database—before generating a response. This approach helps to overcome common issues with large language models, such as generating incorrect information (hallucinations) or providing outdated knowledge.
The system’s methodology involves several key steps:
- Data Preparation: The foundation of Geolog-IA is a carefully curated dataset of 244 pre-degree geology theses from the Universidad Central del Ecuador. Initially, the team considered using unstructured PDF summaries, but quickly realized the limitation: it could only answer qualitative questions (e.g., “What is this thesis about?”). To handle both qualitative and quantitative questions (e.g., “How many theses were published in 2022?”), the data was structured into a CSV file with 16 essential fields, including title, author, year of approval, summary, and keywords. This structured data was then stored in an SQLite database, chosen for its lightweight nature and efficiency in handling text and numerical data.
- Language Model and Agent Selection: Llama 3.1 was selected for its balance of stability, efficiency, and optimization for conversational tasks. For deployment on platforms like Hugging Face, Gemini 2.5 Flash Lite was also utilized. An intelligent agent, built using the LangChain framework, acts as the intermediary. This agent translates user questions from natural language into precise SQL queries for the database, executes these queries, and then uses the language model to transform the database results into clear, coherent, and natural-sounding answers for the user.
Benefits and Impact
Geolog-IA offers significant advantages for various users within the academic community:
- For Students: Provides a clear roadmap for research, helps avoid tedious manual review of numerous sources, and facilitates the structuring and writing of research projects.
- For Teachers: Aids in evaluating student performance, guides content updates, improves teaching practices, and helps orient students toward novel thesis topics.
- For Directors: Offers quick access to valuable information for verifying compliance with institutional objectives and supports strategic decision-making to enhance academic quality.
- For Administrative Staff: Automates information retrieval, streamlining documentation processes and allowing staff to focus on more strategic tasks.
The system’s performance was rigorously evaluated using the BLEU metric, achieving an average score of 0.87. This high score indicates that Geolog-IA generates responses with excellent coherence and precision, making it a reliable tool for academic inquiries.
Also Read:
- Boosting Teamwork in AI: How Prompt Engineering and LLMs Enhance Collaborative Agents
- Collaborative Agents for Automated Data Visualization
Accessibility and Future Potential
Geolog-IA is designed to be user-friendly and publicly accessible. It has been implemented on Google Colab for development and experimentation, and a web-based version is available on Hugging Face Spaces, offering an intuitive interface that can be accessed from any web browser without installation. This ensures that a wide academic community can use the system easily, regardless of their technical expertise.
By democratizing access to geological knowledge, Geolog-IA not only enhances learning, training, and research within the university but also sets a precedent for similar applications in other academic disciplines. This innovative system represents a crucial step forward in knowledge management and information access for the university community.
