Building Trust in E-Commerce: Qubiz's Hackathon Win with AI-Powered Review Validation

The Oradea Tech Hub's HackTech is one of the largest AI-focused hackathons in the CEE. Backed by global sponsors such as OpenAI and Brillio, the event brings together talented coders, AI specialists and industry leaders to unite their creative forces, attend workshops and build innovative AI solutions.

Qubiz participated with two talented teams and three dedicated mentors, each focused on tackling innovative challenges in the AI space. Competing alongside like-minded professionals, they pushed the limits of creativity, technical prowess, and collaboration to win the challenge created by Brillio.

Reviewlio: Shaping the Future of E-commerce Reviews with AI

This interview dives into the journey of Maria, Mihai, Paula, Alex, and George, who impressed the judges with Reviewlio—a browser extension designed to detect fake reviews on e-commerce platforms.

What inspired you to join OTH's AI Hackathon?

Maria: You (Krisz) approached my desk, dropped a challenge before me, and said, "You in or what?" I couldn't exactly say no to that! Jokes aside, it was an excellent opportunity to push my skills and work alongside like-minded people on real-world problems.

Mihai: I have always had an interest in AI, and this, coupled with the fact that I had just completed an AI Workshop, made me excited to apply what I had learned. So, of course, I said yes.

How did it feel to compete with other AI enthusiasts and professionals?

Maria: The organisers created a collaborative yet competitive environment where everyone was eager to share ideas while working toward their unique solutions. It felt refreshing to be among other professionals and AI enthusiasts, all bringing different experiences and perspectives to the table, making it a powerful learning experience.

Mihai: One word: awesome. From the people I met to the people I already knew, and especially to the team I was a part of. Knowing that we were all here to have fun and learn new stuff was something special.

Can you describe the challenge that your team won and what it entailed?

Maria & Mihai: Our team won the Brillio challenge, where we created a browser extension, Reviewlio, to detect fake reviews on e-commerce sites. The Chrome extension analyses various attributes of product reviews to assign each review a trustworthiness score. This score helps users quickly identify which reviews to trust, making purchasing decisions more accessible and informed. In the back, we had a server that scraped the URL for data, a Large Language Model (LLM) to give us an initial understanding of each review, and a machine learning (ML) model to provide us with the final score.

What was the most difficult aspect of the challenge, and how did your team overcome it?

Maria: The biggest challenge was integrating all the components, with team members working in both familiar and new areas and some stepping outside their usual expertise. We had to coordinate to connect everything seamlessly and quickly, and, in the end, putting it all together just in time was incredibly rewarding.

Mihai: Each team member was responsible for a part of the process, and as each member chose something outside of his area of expertise, the most challenging part was putting all the puzzle pieces together. We managed to do that just in the nick of time.

Were there any unexpected obstacles that required you to change your approach?

Maria: Yes, working with the two provided datasets, which only included review text and basic labels (CG for computer-generated, OR for valid).

We found this insufficient, especially since some reviews were short and ambiguous, complicating our ability to assess trustworthiness. This variability made it difficult to create a reliable training and testing dataset for our model's accuracy. To overcome this, we developed a set of questions that guided our investigation into trustworthiness, allowing us to analyse additional features like language patterns and sentiment for a more nuanced measure of authenticity.

Mihai: On my side, which was the integration and communication with the LLM, the unexpected part was getting the LLM to return correct responses consistently. It took a lot of "prompt engineering" for that to happen, and only in the end did I get a consistent response after understanding why the LLM was behaving that way.

Which aspects of your project do you believe contributed most to your victory?

Maria: The aspects that contributed most to our victory included strong collaboration, effective brainstorming, and planning, which allowed us to assign tasks based on each member's strengths while keeping everyone aligned with our project goals.

Mihai: The initial brainstorming session and the fact that we all left our comfort zones allowed us to innovate, have fun, and pursue our ideas.

How do you envision the solution being applied in real-world scenarios, possibly within Qubiz projects?

Maria: Our solution is applied in various real-world scenarios, especially e-commerce, where it can help consumers identify fake reviews and make informed purchasing decisions. In the context of Qubiz projects, this technology could be integrated into client platforms to enhance trust and transparency in customer feedback. Additionally, it could be leveraged in market research to analyse consumer sentiment, providing valuable insights that help businesses refine their product offerings and marketing strategies.

Mihai: I’d say that the tool that we developed can work on anything that has a review section, so industries like e-commerce, restaurants, etc. are prime candidates, but generally speaking, any business that is interested in offering the most relevant information from real users in their review section could leverage this tool.

What was your favourite part of the hackathon, and how did it feel to win?

Maria: My favourite part of the hackathon was seeing our solution come to life as we successfully integrated all our components. Winning felt like a validation of all our hard work and creativity, proving that even under time pressure, we could create solutions with the potential for real impact against fake reviews.

Overall, it was a fun and inspiring experience, and I am genuinely grateful for my teammates and what we built together.

Mihai: My favourite part of the event was seeing the project come to life and our solution present itself as quite viable. We were already thrilled with what we achieved as a team (in two days) even before the winners were announced. The fact that we won further validated that feeling and showed us that even under immense pressure, we still created a solution with a real impact.