Invented by Gelli; Francesco, Dong; Yanfei, Lin; Ting, Zheng; Pingxia, Mangalore; Nithin Navin, Palaniappan; Sathish Kumar, Eda; Chenna Rao, Upadhyay; Rushik Navinbhai

Artificial intelligence is changing how we find, review, and understand information. One new patent application shows how smart technology can help people get quick, clear summaries from large piles of web results. In this article, we will look closely at this patent’s ideas, see how it fits into today’s world, explore how it builds on past work, and finally, explain how the invention itself works and what makes it special.

Background and Market Context

Doing research on people or organizations is not new. Employers, schools, and even volunteers often need background checks to learn about someone’s past. For a long time, there have been two main ways to do this: formal checks using special, private records, and informal searches using public web data. The first is slow, costly, and sometimes hard to access. The second is fast and cheap, but can be messy and incomplete.

Most people today turn to the internet and search engines like Google or Bing to find information. They type in names, keywords, or topics and get pages of results. But this is just the start. Reading through all the links, finding what is important, and putting it together takes a lot of time and effort. People often get lost or overwhelmed. They may miss key facts or, worse, make mistakes because some web pages are out-of-date or just plain wrong.

At the same time, artificial intelligence is advancing quickly. We now have “predictive” AI, which can guess or suggest answers based on patterns in data, and “generative” AI, which can write text, make images, or summarize big ideas based on prompts. Tools like ChatGPT, Dall-E, and Google Bard are becoming common, but using them well is not always easy. If you do not ask the right question or give the right prompt, the answers can be confusing or off-target.

Many businesses, law firms, schools, and even regular people want faster and more reliable ways to search for information and get clear summaries. They want tools that can cut through the noise, ignore ads or junk, and focus on what matters. They want a system that is easy to use, gives the same good results every time, and saves them hours of work. This is where the new patent application steps in.

By mixing predictive and generative AI in a careful way, the patent aims to solve a real problem: turning big, messy search results into simple, useful reports. This is not just for lawyers or investigators, but for anyone who needs to know more about a person, business, or topic. The market for such tools is growing fast as more people rely on the web for research and as the volume of online data keeps exploding.

Scientific Rationale and Prior Art

To understand why this patent matters, it helps to look back at what has been done before and why those old ways fall short.

In the past, background checks were done by people. They went to different websites, read lots of articles, and tried to find the truth by hand. Some companies built tools to help by searching websites and showing results in a list. But these tools did not read or understand the content. They just gathered links. The user still had to do the hard work of reading, picking out what was important, and writing a summary.

Later, some systems tried to sort results by “relevance,” using simple rules like how many times a word appears in a text. Others used basic natural language processing to find names or dates. Even so, they could not really “understand” what was being said, and often mixed in noise—like ads, spam, or unrelated pages.

More recently, “machine learning” models have been used to group or filter search results. These models can find patterns and spot what looks like a match. But even with this, the end user often still has to read many pages and make sense of the findings.

The biggest leap in recent years is the rise of “generative” AI models. These models, like large language models (LLMs), can write short summaries when you give them a good prompt and some data. But what is a “good prompt”? And how do you know which data to give the model? This is a big problem. If two people write prompts in different ways, they may get very different answers—even if they are looking for the same thing. Worse, if the wrong data is fed to the AI, the summary may be off, leading to confusion or bad decisions.

Some early approaches tried to automate this by using set templates for prompts. Others tried to “chunk” documents into smaller pieces and compare them to a question. But these systems were limited. They did not mix predictive and generative AI in a smart way. They did not handle “noise” well, and often required the user to do much of the setup work.

This patent stands out because it combines two kinds of AI. First, a model finds the most “relevant” parts of all the search results—ignoring junk and focusing on what matters. Then, another AI writes a summary, using a carefully built prompt that fits the type of research. This lowers the chance of mistakes, makes the system easy to use, and gives results that are more consistent and clear than older tools.

In short, while others have built search engines, filters, or AI writers, this approach is unique in how it brings together these parts into one smooth flow. It is like having a smart helper who not only finds the best information but also writes it up for you, all in one step.

Invention Description and Key Innovations

Now, let’s break down how the invention works and what makes it special.

At the heart of this system is a “research engine”—a mix of software modules that talk to each other to make the user’s job easier. The user starts by opening a simple screen, called a graphical user interface (GUI), on their computer or phone. This screen has easy-to-use parts: a list of research types (like “check for negative news,” “find awards,” or “general overview”), and a box for typing in a name, company, or topic.

When the user fills in these boxes and clicks “search,” the system kicks into action. Here’s what happens:

First, the system uses the chosen research type and target to build a special search phrase or prompt. If the user chose “negative news” about a company, the prompt might include words like “scandal,” “fraud,” or “lawsuit,” along with the company name and any extra details given. If the user is interested in positive achievements, the prompt would include words like “award,” “recognition,” or “expansion.” The system can pull these words from a database of templates, so the user does not have to know what to type.

Next, this search phrase is sent to a search engine—either a public one like Google or a special one for certain records. The search engine returns a bunch of results: web pages, articles, blog posts, and more.

But not all these results are useful. Some are off-topic, some are ads, and some may only mention the target in passing. This is where the predictive AI model comes in. The system breaks up the results into “chunks”—small parts like sentences or paragraphs. Then, it uses a trained model to turn each chunk into a kind of “fingerprint” (an embedding vector) that shows what it’s about. The system also creates a fingerprint for the research target.

By comparing these fingerprints, the model can measure how “close” each chunk is to what the user wants. Chunks that are unrelated will be far away; those that match are close. The system picks out the chunks within a certain distance and marks them as “relevant.”

To make this process fast, all the fingerprints are stored in a special database called a vectorstore. This database helps the system quickly find matches and ignore noise. If a chunk looks very different from everything else (for example, an ad or a list of unrelated links), it can be dropped.

Once the best chunks are found, the system gets ready to write a summary. This is where the generative AI comes in. But instead of letting the user write a prompt—which could lead to mistakes or confusion—the system builds the prompt automatically. It uses templates based on the research type and fills in the blanks with the findings and any extra details. If no relevant news is found, the prompt tells the AI to say so, avoiding confusion.

The generative AI model, like ChatGPT or Bard, then takes the prompt and the relevant data and writes a clear, short summary. This summary can include a “Sources” section at the end, listing the links used, or any other format the user wants.

Finally, the summary is shown on the user’s screen. The user can read it, edit it, or give feedback. The system can remember feedback to improve future searches.

What makes this invention stand out is how it puts all these steps together. The user does not need to know about AI, machine learning, or writing good prompts. They just fill in a form and get a well-written answer. The combination of predictive (for picking the right data) and generative (for writing the summary) AI makes the process smooth, fast, and reliable.

Here are some unique features:

– The system can handle different kinds of research: from finding negative news to listing awards or general facts.

– It can use different search engines or special databases as needed.

– It breaks up big documents so even small, hidden facts are found.

– It stores and compares results using embeddings, which are much smarter than just counting words.

– It builds prompts automatically, so every user gets a similar, high-quality summary.

– It can adapt based on user feedback, getting better over time.

This invention is not limited to just background checks. It can be used in schools, newsrooms, law offices, or by anyone who needs to make sense of lots of information quickly. The system could also be updated to work with images, audio, or video, making it even more powerful in the future.

Conclusion

This new patent application brings together the best of predictive and generative AI to solve a real problem: turning huge piles of search results into clear, useful summaries. By mixing smart models, easy-to-use screens, and automated prompts, it saves people time and helps them make better choices. In a world where information is everywhere but time is short, tools like this are set to become a key part of how we all do research.

Click here https://ppubs.uspto.gov/pubwebapp/ and search 20250217427.