Structured content gets a voice: Engaging dialogues and expert instructions

Recently, Tridion’s team convened with its partner, The QA Company (TQC), to discuss the joint value proposition of both companies, especially regarding TQC’s QAnswer solution. The event was organized by Elsa Sklavounou, RWS, VP AI Alliances, Global Partnerships.

From the TQC team, we had Dennis Diefenbach (Chief Executive Officer) walk us through the QAnswer solution to demonstrate TQC and Tridion’s joint value proposition in practice. QAnswer is TQC’s core solution that uses AI to answer questions using the content contained in enterprise documents.

Dennis introduced TQC as a start-up founded in 2019 and located in the south of France. Its core product QAnswer is available as both cloud-based and on-premise solution. It correlates with Tridion because it works on top of structured data available in the form of documents. That said, TQC also works with unstructured data and is witnessing significant potential in this area.

Dennis acknowledged the fact that TQC is a relatively young company, but at the same time, it has some big clients, most notably the European Commission (EC). It helps EC to interrogate the data, especially legislation. Other major clients include NASA, JPL and Rockwell Automation.

QAnswer works with structured and unstructured data

Dennis used two different demonstrations to walk us through the solution. The first was about working with a single PDF document in QAnswer, including how it answers data-related questions in such a PDF. The second demonstration included working with multiple documents. TQC strives to deliver the most accurate answers in short response times.

Demo 1: Single document

Dennis walked us through the demo, which included all the steps to upload a single document. The key functionalities include instant playground, knowledge graph, and text and document playgrounds. The document is uploaded in the documents playground.

QAnswer Dashboard

Dennis uploaded a refrigerator’s technical document for demonstration purposes. He asked specific questions, such as how humidity is maintained inside the fridge and the answer was displayed:

Answers from QAnswer are not a mere copy-paste of relevant information

Dennis mentioned that the QAnswer platform dives into the document once the query is entered and scans through relevant information. And then reformulates all the information from across the entire document to form an answer. This means it’s not simply copy-pasting the information from the document but combining different pieces of relevant content and forming a structured answer the way a human would.

Example of an answer delivered on refrigerator technical document

The answer also has the source content captured for the user to go through in case they would like to go through the PDF sections used to build the answer. Clicking on the source (right side of the image below) takes you to the relevant part of the PDF, where the relevant content used is highlighted (left side of the image below).

Answer references and sources in QAnswer

The source is important in most cases e.g., Dennis mentioned it’s critical in the technical documentation of an industrial plant or a risk contract.

Demo 2: Multiple documents

Dennis used a set of documents he sourced from RWS’s website to walk us through how QAnswer works. As a first step, he uploaded all these five documents onto the QA System.

Uploading multiple documents onto the QA System

Once the documents are uploaded, QAnswer indexes all the information contained in these documents. Once indexed, the system is ready for the user to ask questions and deliver answers.

Dennis walked us through the demo by entering certain questions and the system delivered relevant answers. There is also an option for the user to get the answer in bullet point format. For this, the user must type the question followed by ‘in bullet points’ and the answer will be displayed in bullet point format.

The answers are displayed almost instantly and there is no limitation on the document’s type or size. Dennis mentioned that one can upload PDF/s, DOC/s, and TXT/s to the platform. Multiple files of these formats can also be uploaded and questions run on the entire dataset.

QAnswer has an ‘Export’ functionality, essentially a widget-like Chatbot interface. Once the content is indexed and the question answering system is created, it can be exported into this format on a website.

All the QAnswer functionalities can be integrated into a solution such as Tridion through APIs, which works with multiple languages.

Free demo for users

Dennis mentioned that the demo is free for users, whether registered or not, to test the platform. All the playground options are available for testing.

Discussion on the key functionalities and use cases for QAnswer

Dennis then opened the forum for participants to ask questions. Some of the key points discussed were related to the following:

Accuracy: 

Dennis mentioned that the QAnswer doesn’t guarantee 100% accuracy in results, which is also why it shares the sources for the user to go in and check the original content in the document. This answered a question: Can medical companies use the application to understand how the device works?

Large language model:

QAnswer is also using a large language model, but this has been built in-house and serves the customers who want on-premise solutions, so they don’t need to share their data with a third-party solution provider. And it has all the functionalities built on top for more accuracy and quick response time.

Feedback Mechanism:

There is an option for the user to like or dislike the answer. TQC uses this feedback to refine the model to deliver better results. However, the QAnswer tool also has a regenerate button; if a user is not satisfied with the answer, then this can be used using a different approach to generate a new answer.

Crawling the entire website or parts within:

Dennis confirmed that the tool could be used to crawl the entire website or part/s within and then answer questions related to the website / URL.

Other

QAnswer doesn’t have an option where it prompts questions to the users; Dennis mentioned the reason being that there is only a limited field in the underlying documents and hence there is only a little variety of questions that the user can ask.

QAnswer connects to data repositories such as the one at European Commission through elastic search APIs. It takes documents indexed in the elastic search and then, for example, exports those into a PDF file.

Conclusion: Lot more to explore even beyond technical documentation

Elsa explored the possibility of QAnswer working on a client's RWS repository, working with all forms of data and serving answers to the users. Dennis mentioned they haven’t worked on such a use case before and can explore this area by working with RWS to serve RWS’s clients jointly.

Elsa acknowledged that QAnswer bringing in the interrogation model offers much potential and is the next level of search technology. Also, given the RWS’s headless delivery capabilities, it would be interesting to explore which channel/s users would prefer to use for this technology. Elsa believes it will encourage users who prefer ChatGPT to use this on their documents to generate relevant answers to their questions.

As the next step, TQC will work with RWS to test certain immediate use cases that QAnswer can address and explore other use cases later.

If you would like to learn more about how Tridion can help your business, click here.
Learn more about what structured content is.