Hi Michael, tell us a bit about your journey so far and how you ended up at FoodFacts?

I’m originally from Mexico where I studied mechatronics engineering, which is a combination of mechanics, electronics, and computer science. I have a Masters in AI from Edinburgh University, then taught a course called computer vision for robots at the Monterrey Institute of Technology.

I came to Sweden to do a Master's in Entrepreneurship and most recently I’ve been working at an electric car startup building their self-driving algorithms and figuring out how to produce a car that minimizes pollution.

About a year ago I found out that FoodFacts was looking for someone who could come up with a way to estimate the volume of each ingredient in a food product. To develop an algorithm that automates this calculation. And that seemed like the perfect challenge for me where I could use all my knowledge in AI and machine learning, so I applied for the position and now I’ve been working on that specific project on volume estimation since then, along with other things.

Can you tell us a bit about how you developed the volume estimation algorithm and what it is?

The volume estimation algorithm I have developed essentially estimates the recipe of each product down to an ingredient level. Initially, there are some things that we know about a product, like that the ingredient list is written in falling order (from largest quantity to smallest) and we also have the nutritional value available. With this information, we can create equations to calculate the volume of each ingredient backward. 

In an ideal world, we would base these kinds of calculations on a standard industry practice with everyone’s exact recipes, but that world doesn’t exist. Brands usually don’t share their recipes, for good reasons. We are able to create calculations that are good enough estimates of the volumes of the ingredients which opens up a lot of possibilities. 

This is really useful when we are automating estimations of the carbon footprint of a food product. Each ingredient is given its own carbon emission footprint, and we are able to avoid using complex and expensive life cycle analysis and can still get a really good estimate of the carbon footprint. It democratizes the availability of carbon footprint calculations, products can be compared in a fair and science-based way, enabling new digital services or better business decisions. 

It has been a long process to develop this, it’s not only to come up with the equations but also which programming library to use to solve complex math problems, how to incorporate constraints and generate constraints using computer code. And debugging the algorithm takes up most of the time. But I’ve been able to use my experience in both machine learning and AI which has been super valuable along the way. We have also sense-checked our hypothesis for the carbon footprints with expertise from Stockholm Resilience Center to look at the impact of packaging and combined it with data from RISE to increase the accuracy of the automation.

What food fact do you think everyone should know?

Food scarcity is actually only a problem of logistics. We produce enough food to keep everybody in the world well-fed, but we throw away about a third of what is produced. This includes food waste at the production and distribution stages, as well as food that is thrown away by consumers, restaurants, etc.