How I built a 220k $ app in 4 weeks

My name is Dragos Dobrean and I’m the co-founder of appssemble — a mobile development studio, besides that, I’m also doing my PhD on Mobile Applications Software Architectures, you could say that I live and breath mobile. I built an app that counted the number of calories in your meal by only snapping a photo of your food (aical) and got a signed offer 100.000$ for 45% of it in 4 weeks.

So how did I do it? Everything starts in Cleveland, I was attending an academic conference (IEEE International Conference on Software Maintenance and Evolution) where something caught my attention, there was a talk by Alex Cummaudo where he presented his research on AI/ML API’s and did a comparison of those (Google Vision, Amazon Rekognition, IBM Watson, etc.).

After a week or so, I was thinking one evening that I could leverage those services, to build something cool. All of those services offer image tagging and labeling functionalities via an API (basically, you feed them the picture, they give you back all the information regarding that picture). I saw that those services were good at recognizing food from images, so with this idea in mind, I started thinking about building a calories counting app, that would work by simply snapping a picture of your food.

Building such a product from scratch is a tremendous amount of work, you would need large data sets of food images to train your model, validate it, and pack it in a form that could be used from a mobile device, fortunately, I did not have to worry about this kind of thing, as the model was ready and it had word-class accuracy, I mean few companies are better at AI/ML than IBM, Google or Amazon.

Now I had the labeling part working, but how do I go from “french fries” to 319 kcal. There was one missing bit, converting a food name to the correct number of the calories, only if there was an API for that, well, fortunately, there is, after a couple of hours of search the web I found Edamam a company which does exactly that, providing an API for nutritional purposes.

This was the first week, I hadn’t got anything built yet, but I had a pretty good idea of how everything is going to work. The second week came by and I started coding the app, firstly I decided on how I wanted it to look, I wanted something minimal but eye candy so this is what I came up with.

The second and third week was spent on coding the app, connecting the APIs, and creating the user interface. By week 4, I’ve prepared some marketing materials, and I released the app on ProductHunt. The app had major success, it had over 2.000 downloads in the first week.

By this time, I’ve received an offer from an investment firm that was interested in acquiring 45% of aical, and after exchanging a couple of emails, they have already signed their end of the deal.

However, I didn’t get rich as I haven’t taken the money. The reason for that being that it didn’t seem fair to do that, as the app could not be further improved, the app would work fairly well on some of the cases but the major issue had been estimating the size of a meal and choosing the best labels.

I’ve tried to overcome this issue by allowing the users to choose one of the most convenient labels and enter their quantity, unfortunately, this didn’t have too much success.

To improve the accuracy, I’ve also implemented a Siri shortcut that counted the calories using voice commands, while this had some popularity with the users, I think the timing was not good for this, as the users are not that acquainted with using voice commands yet.

What happend next?

Not much, the app continued to grow, it has over 12.000 downloads, active users, and even paying ones. The app is really big in Saudi Arabia, but I haven’t found the reason for this, maybe someone in the comments can shed some light on this…

Now a few learnings from all this process:

• You can build something really interesting by relying on others work, you don’t have to be Google or Amazon, but you can leverage their services
• A product doesn’t have to be perfect for users to use it
• The accuracy of aical was really good, there are startups with a lot of investment (like FoodVisor) that don’t achieve this kind of accuracy on their own

That being said, you want to find out more about what I’m about to, follow the blog, or reach out to me.