As more and more organizations are looking to adopt AI, it is becoming increasingly important to find the right balance between using public AI models while protecting private data and access. The XFinLabs mi-AI platform is designed specifically to address this issue. Companies are able to quickly and easily integrate AI on-prem or on-cloud to safely and securely access internal data while also getting the full benefits of AI.

The XFinLabs mi-AI platform

XFinLabs mi-AI allows you to use any LLM to automate and generate responses to questions that people have on a day to day basis. From the mundane such as “Where is waldo?” to the more complex such as “Analyze the last four years of our balance sheet and itemize discrepancies across each division by product. Generate a report, from largest to smallest.”

Under the Hood

Prompt Engineering

Automated Workflows

Hyper Tuning

XFinLabs mi-AI is a Retrieval Augmented Generation (RAG) based AI system. RAG systems are ideally suited for enterprise environments as it allows companies to control what data sources their employees use. In the diagram below, an employee uploads their data into mi-AI. (This data can come from a variety of sources such as files, company databases or even the web.) A unique database is then created that only stores the information provided by the user. (This helps secure company data as well as ensure the accuracy and relevancy of any AI responses.) Once created, users can use the power of semantic search to query their data as well as automate specific tasks and operations. With XFinLabs mi-AI, a company can have as many private AIs as they want.
A critical component of the XFinLabs mi-AI system is the ability of a company to tailor how the AI is trained and responds. This is often referred to as the process of giving your private AI a personality. For example, with mi-AI you can tell your AI that it is a domain expert in certain technologies and should respond to queries in a professional tone, as opposed to a more casual tone. You can even narrow its core competency to a specific topic. Such as informing the AI model that the data contained in the private AI database is only related to "the industrial revolution circa 1800 to 1840". Once these prompts have been entered into mi-AI, the next time a person does a search, anything outside of the above domains will either be ignored or referenced as immaterial. Because of this flexibility and granularity, (and potential for misuse) prompt engineering should only be carried out by company executives who can control exactly how their AI responds based on the business they are in.
One of the biggest challenges in leveraging AI is in understanding how to use AI to discover trends, patterns and anomalies. Tasks that analysts can spend countless hours, days and weeks trying to identify. Rather than asking your private AI about employee salaries, perhaps a better use of the tool is to ask, “are our employees being fairly compensated based on market rates and what competitor A and B are offering.” This is where AI excels, but it requires users to understand the best way to first frame a question and secondly, know where to go to find the facts. To help, XFinLabs mi-AI provides Automated Workflows where users can point mi-AI to pull data from internal and external sources, searching for specific words or phrases and then running an analysis at a specific time or on a schedule. This feature is particularly useful when doing market research or engaging in due diligence. With Automated Workflows analysts are now to have their private AI churn through 10ks, read press releases or catch mentions of potential transactions or events that may materially affect a business, product or market.
One of the most powerful features of RAG based systems is there ability to hyper tune the AI model. Hyper tuning affects the underlying algorithms to obtain more accurate or relevant responses. It is much like adjusting the knobs on a radio to get the clearest signal. For many, the sound that comes out may be more than adequate, but for the audiophile (in our case an analyst) who can pick up disturbances or static, the ability to find the right wavelength ensures crystal clear sound. For AI, this means better results that a tailored specifically to the user and/or organization. As expected, with this level of control comes some responsibility. If hyper tuning a model for a company, the user has to be well versed in the type of data that private AI model is looking for. This is another key reason why XFinLabs mi-AI was created. It allows a company to create one private AI with it’s own personality and frequency (hyper tune), while another private AI (looking at the same data) may be configured with a different personality and level of tuning.