Beltone Holding concluded its first AI hackathon earlier this month, challenging students to roll out their own AI solutions predicting gold price fluctuations, according to a press release (pdf). The hackathon was held in collaboration with the American University in Cairo’s Mathematics, Actuarial, and Data Science Association. To learn more about the hackathon, EnterpriseAM spoke with Beltone Holding’s Chief Data Scientist Basma Rady (LinkedIn) and Hesham El Alamy, one of the students who participated in the event.

A big turnout: Over 400 students from the AUC and the German University in Cairo took part in the event, with 30 teams submitting over 450 solutions.

EnterpriseAM: Could you tell us more about the hackathon and what encouraged you to hold it now?

Basma Rady: The hackathon is a unique chance for students to get hands-on and real life experience with data science and AI. We wanted to give as many students as we possibly can the chance to get their hands dirty with data science and help them break into this world. We already take in six interns every year, showing them what the field looks like and how it’s applied — but we also take them through the most important steps for building an AI model and applying it to the financial or any other sector. The hackathon is meant to give this chance to more students.

The structure of the hackathon was designed in a way that would highlight the importance of technical skills, staying up to date with advances in data science and AI, being able to use cutting edge tech to develop a solution to a very hard problem — and the importance of being able to explain it, regardless of whether or not the audience comes from a technical background.

The hackathon spanned a full month. The problem statement was: “can you predict the percentage change in gold prices in Egypt?” The data set that we curated for the hackathon was in itself something innovative, as it combined numerical and textual data from Egypt and global markets. The data included things like the monthly inflation rate, stock prices, the Fed’s interest rate, daily news, and the intraday gold prices in Egypt, among other things. We then asked the students to predict the difference in gold prices in Egypt for the next day.

The teams that submitted the best performing models were invited to the closing day, where they would submit their findings to a jury of technical experts as well as people without a background in data science. The non-technical people would ask the teams to explain their models — and to do so in simple terms, in a way that someone from a business background would be able to understand. Even if a certain team had the best model, it didn’t end there — they had to explain it, present it, and be able to convince the business side that this model was worth their time and attention.

E: What did the winning team present? What did their model look like? And why was it chosen?

BR: The winning team had a solid model and its members were able to explain their findings in an impactful way. They provided compelling data visualization, explained the problem statement, and they had developed a model that predicts the percentage change in gold prices.

E: How is Beltone going to benefit from the winning model or other models produced during the competition?

BR:We were inspired by a few ideas from the top teams. While they are still student-built models, they still gave us inspiration for new ideas that we could develop further. From Beltone’s side, a part of it is about fostering data science talent that is able to take academic concepts and apply them to real world problems.

E: How might such initiatives bridge the gap between academia and organizations?

BR: During the closing ceremony, our group CEO asked the winning teams what they learned during the hackathon — and the common sentiment was that they had to learn to work with a deadline and to manage their time properly. Another point of relevance was showing them what a data scientist would face in real life. Finally, they had to learn how to sell their solution — how to convince the most technical people in the room that the model works, and also convince the business people in the room that they should trust their model.

E: How are AI applications being applied in the financial services sector?

BR: AI is used across the board, be it for generating new content, or even for credit scoring and with coming up with new products for clients. The work from our research — on the use of machine learning and AI in order to forecast volatility for stocks in the EGX — and the output from the hackathon are currently being merged with and leveraged by our asset management platform in Beltone. We’re also doing a lot of interesting stuff with language models and news data, and extracting sentiment from news, categorizing it, and being smarter with how we consume news internally.

E: Can we expect more hackathons from Beltone soon?

BR:We’re hosting another hackathon this year, and next time it won’t be limited to the AUC and the GUC communities. We want to collaborate with Ain Shams, Cairo University, Alexandria University, as well as new universities in the new capital.

E: Do you plan on any more planned collaborations with the talents that emerged during the hackathon?

BR:We’re in touch with the top six teams, and we’re planning on inviting them to shadow us and show them how we’re applying data science, and we’re also going to have them sit with our asset management team so that they can get them to see the business side of our world. The winning team will also intern with us for a month.

WHAT ABOUT THE STUDENT SIDE-

EnterpriseAM spoke to Hesham El Alamy, one of the students who participated in the event, for his take on the hackathon.

E: Can you tell us about the solution you presented?

HEA: The hackathon was about predicting the percentage change in gold prices, and we were given real life data sets that were gathered by Beltone. It was very challenging compared to other competitions and projects that we took part in. We had two weeks, and we thought we needed to have a strong model in terms of the technical side and a strong machine learning model, but we also wanted to create a comprehensive solution that’s built on statistics, economics, and business solutions for the end user.

E: Tell us about the solution you came up with.

HEA: Our machine learning model utilized a new idea that we developed called the Stacked Regressor. We used our predictions and the data sets that we were given to build an automated dashboard that can help Beltone employees find relevant data that they could analyze easily. We were given structured data, but this dashboard can help you find insights, analyze the numbers, and understand how this data was gathered, helping them use it in their daily operations.

The most important takeaway from the hackathon is that even if you have a really strong model, it won’t be appreciated if you can’t sell it right and be able to link it with business solutions. An understanding of the business side that coincides with an understanding of machine learning is needed.

Your top education stories for the week:

  • EBank launches InfoSec Academy for IT grads: EBank, in partnership with the CBE’s Egyptian Banking Institute, has launched its Information Security Academy to equip university graduates with skills in cybersecurity and IT. (Statement | pdf)
  • Egypt plans to add 1k schools: The General Authority for Investment and Educational Buildings Authority aim to attract investments to establish 1k schools by offering fresh facilities.
  • Social Impact Capital submits offer to own up to 100% of CIRA: CIRA Education majority shareholder Social Impact Capital (SIC) has submitted a mandatory tender offer to acquire up to an additional 48.78% of CIRA. SIC currently holds a 51.2% stake in the country’s leading private education company. The offer is currently being reviewed by the FRA.
Tags: