Presenting AI-Powered Chat With Your Documents

In today’s business environment with massive data sources, employees struggle to find the right information in the shortest time. Otherwise, they cannot solve problems efficiently. To address these concerns, IBA Group developed a solution called iDocIt.

How did the idea of creating iDocIt come about?

As a developer, I was trying hard to solve a problem and could not find the answer in the documents. Then I came up to more experienced colleagues. They rubbed my nose into the manuals and the solution was there. For some reason, I overlooked this portion of the manual. My problem was that I did not formulate the question correctly when searching across the documents. Other colleagues had similar or even bigger problems. Documents can be stored not only in different files or folders. They can be stored even in different platforms, so one should know where to look for the answer. Eventually, I decided to automate and simplify the search process, and thus help not only myself, but also others.

What problem do users have and how do they solve it now?

Though they talk a lot about Generative AI, numerous companies have been working with the products that are 5-10 years old. There are enormous arrays of policies, manuals, databases. At best, it is possible to search using key words. Typical search gives a list of documents where the information you are looking for may reside.

Today, there is a technical capability to solve the problem. For instance, our iDocIt gives an answer to the question and links to the documents from which the answer was extracted.  If you want to clarify or to check, if the answer is correct, or receive an extended answer, you can follow the links. I would like to highlight that iDocIt gives the answer and not a link to, say, a 12-page document.

What distinguishes the new product from those already on the market?

Unlike the existing products, iDocIt allows the user to upload the entire knowledge base of a company. As the answers are not based on some generic information from the web, they are precise and valid. Our approach minimizes the risk of hallucinations, cases where AI generates inaccurate answers when they are not available in the knowledge base. We adapt the inquiries that we send to the LLM (Large Language Model) and use the approach called Retrieval Augmented Generation to minimize hallucinations and increase precision. We slip the LLM a cheat sheet. Here is the question and here is an extract from the relevant documents. We split the entire knowledge core into categories to avoid confusion because a similar inquiry and answer may have a different context. Thus, we improve the quality of the answers.

What innovations does iDocIt offer?

iDocIt can do search both in fixed documents and in real time data like, for example, Microsoft Co-Pilot searches information in the web and then compiles an answer. We can search not only in the web, but also across internal systems, making an inquiry through an API call and then generating an answer. How do we know we need real time data? It depends on a category. For example, the Latest News category assumes that the application needs to look for an answer in a specific website.

How should the customer use our product to get the most value from its use?

To maximize the benefits of iDocIt, customers should have a Help Center in place. Without a Help Center, it is difficult to measure outcomes and track efficiency improvements. Typically, 80% of user inquiries fall into 20% of categories with standard, well-known answers. These questions can often be answered without human involvement. Implementing iDocIt to handle these queries can yield significant efficiency gains.

One common concern with public LLMs is the need to send sensitive data to the cloud. However, iDocIt offers the option to install the LLM on-premises, ensuring that all data remains within a secure perimeter.

Who are the potential customers of iDocIt?

The most promising clients are large-scale enterprises that want to have a help center or those that already have a help center, but want to optimize its operations. Another group is companies that have many clients and need to advice the clients on its products a lot. These may be banks or insurance companies with numerous insurance policies.

In a broader sense, we target iDocIt at companies that have an extensive base of structured and unstructured documents, seek for effective ways to gather insights from their data, and look to provide more informative and responsive services to their clients. We also encourage all our readers to use iDocIt to enjoy an AI-powered chat with their documents!