All Tech Field Day Events

Qlik Answers: Swapping out Generic Results with Smart, Curated Responses

As generative AI is finding its way into the heart of every business, the store of the future looks different. AI tools are transforming industries, boosting productivity and profit at an unseen scale. But experts raise alarms about using the technology without checks and balances.

Turns out, GenAI has a dark side, and it is riddled with errors and hallucinations. Large language models tend to spew out wrong and made-up answers when a question is outside its scope of knowledge.

At Qlik Connect 2024, Qlik announced Qlik Answers, a brand-new GenAI solution that puts an end to the problem of hallucination. It’s one of Qlik’s new self-service AI solutions from the Qlik Analytics stack.

GenAI Has an Achilles Heel

It may be that we have been using the GenAI technology wrong so long. Contrary to what many believe, it is rather simple to prune the errors, and tune an LLM to give back correct answers. It just needs to be directed to good data.

“A large language model as a stand-alone piece of technology is not an enterprise-grade GenAI solution,” says Ryan Welsh, field CTO for Generative AI at Qlik, debunking the most prevalent myth.

Presenting Qlik Answers to the audience at Tech Field Day Experience at Qlik Connect 2024, he explained the other parts which make GenAI whole.

LLMs can be directed to sift through massive volumes of information and answer questions from the data. But they tend to perform at a suboptimal level when the data used for the job is low-quality and stale. They can’t help but get things wrong, and to overcorrect, they fabricate information.

“We’ve found that people typically spend about 20% of their time looking for information or answers in unstructured text data. If you actually calculate that, that’s about one day a week,” he points out.

In an enterprise scenario, that dominoes into project delays, operation slow-downs, and poor content management, among other things.

Results and Answers, Not the Same

Old-fashioned search bars are still around to help find information from large datasets, but the results they pull up are generic, a far cry from the smart, targeted answers that mainstream GenAI tools whip up in seconds.

“Traditional search is broken,” Welsh says. “There is complete difference between a search result and an answer, and now everyone who has had the experience of using ChatGPT understands that.”

GenAI seizes the mantle by delivering the most eloquent phrases and informative narratives that people now have a taste. Only there are perpetual flaws in them. Some errors aren’t so obvious, as Welsh demonstrated through a sample ChatGPT response to a question that was outside its scope of knowledge.

“[It’s] completely unacceptable that you have software in an enterprise that generates 10 to 20% of answers just completely made-up.”

But wouldn’t it be much simpler if chatbots could just say “no” to questions they cannot answer?

Experiments show that when LLMs are trained to use information from trusted sources, they can answer questions with elevated degrees of accuracy.

But good data alone won’t cut it. The team at Qlik knows that to get LLMs to provide highly personalized and fully accurate responses that are cut-out for respective users, GenAI needs something more. Implementing Retrieval Augmented Generation (RAG), an NLP technique that makes relevant corpus of data accessible to LLMs, provided that missing piece.

“We believe every search bar in every enterprise is going to have RAG or an answer engine behind it at some point in the next three to five years,” Welsh predicts. “We’re seeing a massive switch from search bars that deliver a list of results that users need to click through, read, and Ctrl + F their way to the relevant portions of the text to find the answer, to just being able to ask natural language questions and get personalized and direct contextually relevant, trustworthy answers.”

Qlik Answers: Customized and Correct Answers Always

Head of AI practice, Kyle Jourdan, demoed Qlik Answers to the delegates. It is a simple plug-and-play solution that puts a smart, AI-augmented, knowledge assistant on top of any application. Just ask it any question from the domain-specific unstructured data it is directed to use in natural language, and within a flash, it types out a perfectly personalized answer.

“This tool lives where the people are looking for the information, whatever that might be. It’s embeddable in all different places,” Jourdan says.

For a super-smart application, Qlik Answers has a simple architecture. At one end, are enterprise connectors or data sources that it ingests data from. Users can create connections by selecting data sources, or import files directly from Qlik Catalog. The accumulated data serves as the knowledge base for the model. You can expand it by bringing in more sources and datasets as you go.

Before the data turns into insights, it is indexed in the background where datasets are broken down into chunks, embeddings are performed, and stored away in Qlik Cloud.

“All that happens in the background in a matter of a couple seconds,” he says.

At the opposite end, APIs push out results displaying them in the apps it is embedded to.

Qlik Answers’ superpower is not its 100% accuracy, but its ability to say no when required. It only works when it has contextual data to work with, Jourdan confirms.

So when a question is outside its scope of content, it plainly replies that it doesn’t have the answer.

“What we didn’t want to happen is for someone to go ask a question and get the wrong answer from the internet that is not relevant to the policies of that organization.”

Qlik Answers references all results for maximum reliability. Sources are displayed in a separate tab in the same screen.

The search history affords users granular visibility of the prompts and answers that introduces scope for improvement. A feedback system lets users leave a review of whether or not an answer was helpful.

Qlik Answers will be made available to the public July 30th onwards.

Be sure to check out Qlik’s presentations from Tech Field Day Experience at Qlik Connect 2024 to get into the weeds of its self-service AI stack.

About the author

Sulagna Saha

Sulagna Saha is a writer at Gestalt IT where she covers all the latest in enterprise IT. She has written widely on miscellaneous topics. On she writes about the hottest technologies in Cloud, AI, Security and sundry.

A writer by day and reader by night, Sulagna can be found busy with a book or browsing through a bookstore in her free time. She also likes cooking fancy things on leisurely weekends. Traveling and movies are other things high on her list of passions. Sulagna works out of the Gestalt IT office in Hudson, Ohio.

Leave a Comment