Applying AI to Solve Large Data Problems

TTGI's solutions is a standalone system which will have an ATO applied to allow running on the end userĺs existing computing resources. The system will use secure containers, leveraging the Air Forceĺs Iron Bank registry of hardened containers where available to reduce supply chain risk.áThe system will be built around Open Web-UI, allowing conversational interface for seamless human-computer interactions. Initial prototypes allow the user to drag and drop data files into the user interface and Open Web-UI uses Tika as the content extractor, Ollama for embeddings, and storing the data in a ChromaDB enabling Retrieval Augmented Generation over the user provided document set. Gathering, and cataloguing the large quantity of data available to these potential end users creates a foundation for Digital Transformation and other modernization tasks in addition to using local inferencing to generate information from previously underutilized data.For the AI engine, the initial application will be developed around the Ollama LLM server. Ollama was chosen because it is available from Iron Bankĺs registry and allows for an agile maintenance process as upgrades become available (and warranted) - helping work through the ATO challenge to get this AI application in the hands of new users.áOpen Web-UI's ability to stand alone, in hardened containers, allows our target customers to control resource usage while keeping data in their existing secure environment. Using resources already available to the user will limit the cost of using the application, avoiding the unknown fees of using cloud-based resources. Our solution will enable fine tuning models and developing agents that will improve the outputs from the AI Application - tailoring each deployment to the specific needs of that customer.The initial deployment of TTGI's solution is designed to get otherwise disinterested users applying an AI solution to a problem we know exists. Demonstrating the value of AI will open the door to the larger adoptions, Fine Tuning Models and developing Agents. The target customer sets have extremely large quantities of data, including very technical data. Developing an automated pipeline to ingest data and fine tune LLMs can improve the relevance of LLM responses. Adding fundamental knowledge to the application, such as basic physics and various disciple specific fundamentals will also help provide a more advanced starting point for the LLMs. This fundamental knowledge and other data sets can be accessed via agents developed, vetted, and placed inside containers ľ and integrated with Open Web-UI. Ultimately the architecture of this entire system is designed to be scalable, easy to integrate with other applications and agile for continuous expansion of capabilities - but with offramps for users only interested in the basic functionality to solve their initial data problem.á