Lumina AI is a local, privacy-focused AI assistant designed specifically for college students, by a college student. Lumina leverages a Retrieval-Augmented Generation (RAG) model powered by LLaMA (Large Language Model Meta AI) to provide intelligent, context-aware responses and insights based on user-submitted documents. Additionally, Lumina AI includes a database AI agent that analyzes, summarizes, and connects documents, making it a valuable tool for students and researchers.
GitHub:
LuminaAI - Ray Poulton
In the fast-paced world of information, students are often overwhelmed by the sheer volume of course work they must manage—ranging from research papers to math tests and large lectures. Effectively navigating this complexity is essential for acing tests and getting a degree, yet the time and effort required to analyze and extract insights from these lectures/notes can be a significant burden. Enter Lumina AI, a local, privacy-focused AI assistant designed specifically for college students. Built on a Retrieval-Augmented Generation (RAG) model powered by LLaMA (Large Language Model Meta AI), Lumina AI provides intelligent, context-aware responses and insights based on user-submitted documents. Lumina AI is solo developed by me from my college dorm room, and what sets Lumina apart is its AI Database Agent, which goes beyond simple retrieval to analyze, summarize, and connect documents, uncovering patterns and generating actionable insights. By operating entirely locally, Lumina AI not only ensures that data remains secure and private, but it can also be used from anywhere with no Wi-Fi required. Whether summarizing long slideshows, identifying connections between research papers, or extracting key takeaways from English notes, Lumina AI empowers students to max out their studying in a fraction of the time. With its ability to automate time-consuming tasks and provide intelligent support, Lumina AI is poised to transform how efficiently college students interact with data, enhancing productivity and most of all, GRADES.
Lumina is built on a robust architecture that combines natural language processing, document retrieval, and generative AI. The foundation of the system is a Retrieval-Augmented Generation (RAG) Model, which retrieves relevant information from user-submitted documents and generates LLaMA AI responses. The user-submitted documents are split into chunks using a RecursiveCharacterTextSplitter, which ensures the documents are formatted through the AI thoroughly. The unique feature of Lumina is the Database AI Agent, which makes connections between documents, generates summaries of the documents with insights/feedback, and then stores the insights back into the database for seamless retrieval and integration with the RAG model. Both the RAG model and Database Agent are utilized through LangChain to chain together a response memory for more accurate and contextual insight. Additionally, this system runs completely local on the user's machine, which is extremely useful for college students who need to work from multiple locations with no Wi-Fi. The documents can be submitted into the database in various formats (PDF, DOCX, TXT, CSV, PPTX, and video files). The system extracts the text in video frames using OCR (Optical Character Recognition) and transcribes the audio using OpenAI's Whisper model. Every file is embedded using OllamaEmbeddings before being stored in the Chroma database. The interface is implicated using Flask for a simple, but informative user interface, and is updated continuously for chat with the RAG model and Database Agent.
Many experiments were conducted, and are still being conducted to optimize the performance between the Database Agent and RAG model. However, the biggest test was measuring the systems ability to ingest and summarize multiple documents of completely different file types, formats, and lengths. This aspect is crucial to ensure that the RAG model and Database Agent are able to give the user relevant/helpful responses without getting confused with the data or question.
The system managed to accurately ingest all required types of documents quickly, including large power points. Additionally, the RAG model was able to reference each document after some fine-tuning of the search relevance score. The Database Agent worked seamlessly with the RAG model and even produced quicker response times for the generative responsive from the RAG model. This is because the Database Agent takes the workload off of the RAG model by sending the possible information it will need directly to the database autonomously. Overall, the system of a Database AI interacting with a RAG model produced unexpected, but extremely efficient results.
Lumina AI is a powerful, privacy-focused AI assistant that addresses the unique needs of college students. By combining a RAG model with an AI Database Agent, the system provides accurate, context aware responses based on the user's context and needs. The experiments demonstrated that Lumina AI is highly effective at document ingestion, summarization, and retrieval, with a strong focus on data privacy and offline functionality by running locally. Through all this research and advancement, Lumina has only been in development for a few weeks and is still being experimented with each day. I am also learning how to best adapt this software to solve more real-world problems along the way. Ultimately, Lumina AI doesn't have to just be optimized for students. While students could benefit from this software greatly, Lumina can be optimized for really any industry which involves data such as, education, research, business, engineering, healthcare and much more. The possibilities for Lumina are endless, and there is a high market demand for making AI tools easily accessible to the public.