“I’ve kept my promise!” Since his departure for Malaga in October 2022, the former Microsoft country manager has returned to Luxembourg once a month. Starting next year, Candi Carrera is likely to return even more often--probably to the Luxembourg House of Financial Technology (Lhoft)--to develop the company around his unique product, VingeGPT, a combination of “Value Investing Next GEneration” and “GPT.”

The real story began 25 years ago. Carrera was interested in value investing, devouring books on the subject and everything that the stars of the field had managed to spread on the internet. Then came covid, and the Spanish-born geek decided to turn it into a course, which is now followed by 7,000 students on three platforms. “Until now, I’d give them an Excel spreadsheet and they’d have to look up a dozen pieces of information in company financial reports to calculate intrinsic value. If you had any training at all, it would take you an hour. But then…”

OpenAI (the company that develops ChatGPT) on 6 November 2023 announced that it was going to offer custom GPTs, in other words, tools that you or I could co-develop with it. Five days later…

“Here, in three or four prompts, it was almost instantaneous and in many languages.” Together with co-founder Adriano Bidoli, the two men tried their hand at this new toy. “At first, it was a disaster! We decided to take a course in prompt engineering, a skill that we had to acquire. The first three months were pretty complicated. The results weren’t good,” he admits. But the pair persevered.

Today, the tool is ready. “In three quick clicks, you can compare, you have the share price. We have daily data for 43,000 companies. It’s a tool for equity investors, and only for shares. For the moment, it’s not for bonds or ETFs,” even though the question of offering this service for exchange-traded funds is on the roadmap.

As he talks to you, Carrera can’t resist. He chooses “Nike,” types in his prompt, invites you to discover the results, and is amused that his spelling mistake, made in a hurry, doesn’t hinder him from obtaining a good result.

“When you ask , it gives you three results because there are three methods of calculating intrinsic value: absolute, develop discount model discounted cash--which everyone uses--and discounted future earnings. It has been trained and configured to display the parameters and assumptions it uses, such as 3% growth, 7% cost of capital and 30 years. But these three assumptions are only indicative. We use the cost of capital for all industries published each year by . If he says the cost of capital for industry X is 9.11, I’ll ask VingeGPT to recalculate the intrinsic value using that figure.”

Almost 115,000 financial reports digested

The first version of VingeGPT responded in 30 seconds, whereas it takes 130 milliseconds for a Japanese person (the fastest) or 600 milliseconds for a Dane (one of the slowest). The two co-founders reconfigured a back-end in the cloud in six weeks. “I have to admit, I’m very happy with it. And so are our users.” The product’s ambassador programme is already making waves in India, Mexico, Colombia, the United States and the UK. “We’re growing linearly, not exponentially yet, but we can see that people are using it every day.”

For the moment, the small team is selling training courses on how to use VingeGPT properly. “A new course is coming out because we’ve just finished analysing 20 years of financial reports. 113,612 financial reports representing around 900 American companies and 125 international companies listed in the United States. Each time, we have around 40 data points on aspects of the potential for manipulation, polarity and subjectivity of the reports. We’ll be updating this every quarter. You need a ChatGPT plus subscription to be able to use the full power of VingeGPT. Then we’re going to create a mobile app where people can reflect their portfolio. So either their investment portfolio, or a watchlist of companies that they like and about which they can very easily access the necessary information. The idea is to have a bit of a one-stop-shop for all this.”

The Meluxina advantage

“Lastly, VingeGPT entreprise,” he adds. “We’ve already had requests for asset managers, for banks, for private banks... When a new analyst arrives in a company, the managers can facilitate his or her work. Instead of buying reports from analysts, you can perhaps go faster and better…”

But, in fact, why register his company in Luxembourg rather than in sunny Malaga? “It’s so much simpler than in Spain! Our lawyers were telling us that we should already have processes in place for employees before we recruited a single one! Fixed-term contracts are only possible by derogation and in certain industries, such as catering or hotels, and you need supporting documents... And then Luxembourg has another advantage: the possibility, for Lhoft fintechs, of accessing the Meluxina supercomputer!”

This article was originally published in .