(Photo by Brady Dale)
Tracking what you eat is a pain, but it’s good to do. Logging makes a person more conscious of what they’re eating. Which, in turn, makes their choices a bit more responsible. If you’re watching your weight, keeping a record is a great place to start.
One local dev has developed a system, Fitmeal, based entirely on broad, simple text messages. If you just text it what you eat, without worrying about the details, in a few weeks, the machines can start giving you real insights.
That runs against the natural tendency here, in the age of data obsession: once you start to keep track, why not go all the way to making it exact? Why not track just how much? From which brand? With caloric and nutritional precision?
One reason not to go all-out quantified self, one local dev found, is that it’s just really tedious.
“I don’t think it’s very well made for casual people, Georges Duverger, a developer who lives in Boerum Hill who works at eBay told us in a conversation last Friday. He got interested in the quantified self movement after moving to New York City from Montreal in 2010. As he played around with different decision-making apps, he was learning about machine learning at work.
“I was more interested in the idea of nutrition science itself,” Duverger said, “It’s perfect for a software engineer because you have all those numbers you can track.” Duverger was born in France and grew up in Grenoble. They have a very different food culture in France, Duverger acknowledged.
Duverger knew he couldn’t stand to enter the information himself, so he first tried farming the task out via Amazon’s Mechanical Turk. The system allows users to hire anonymous people to do simple, repetitive tasks, at a stated rate per task. He said he only spent about $50 total on the experiment, and found it put him in an uncomfortable ethical quandary, asking himself whether he was paying people enough for these slivers of time. It worked all right, but he found that there was too much variation in the inputs.
Then Duverger wondered if the calorie counts actually even mattered.
With enough data, could machines effectively infer information about foods based on the words from the meals he tracked and the weights he logged for himself each day?
The short answer is: yes, they can.
The longer answer is that it takes a while to start getting useful insights. He’s been doing it for a while now, so the system gives him good data. It’s open for others to try, but users should wait a good two months of texting in foods daily before they expect to learn anything, Duverger said.
With enough time, it does get interesting, and it very much becomes bespoke to you and your habits.
Look at some of Duverger’s most important words. It weights foods relative to each other, based on their impact on his weight. So a number on the “GAIN” side indicates its weight against other foods for increasing weight. “LOSS” is the opposite.
No surprise here: “Burger” is strong on GAIN. “Cucumber” is strong on LOSS.
There are surprises, though. “Pinot” is associated with weight loss — for Duverger, at least — while “Malbec” falls on the GAIN side. Don’t make any plans for yourself around this information: this probably has something to do with where and what Duverger, in particular, tends to pair with those wines. That said: that’s a more powerful insight than the straight calorie count of each glass (which is probably about the same).
Check out a much longer list of words on his GitHub.
Duverger says he is exploring incorporating texts about exercise, Foursquare check-ins and he may work on a hack to get it up and running more quickly for new users.
“I gave myself a goal and I got it,” Duverger told us about his personal results, “but I don’t know how much that’s a result of a tool and how much of it is paying attention to what you eat.” Does it matter?
Duverger moved to New York City to work for Hunch, a startup that helped people to make decisions, which was then acquired by eBay. Christina Mercando, founder of Ringly, is also a Hunch alum. He started exploring these ideas at New York’s Quantified Self Meetup.
Last summer, we dug into the broader French invasion of Brooklyn tech. Duverger lives in Boerum Hill.
Check out the 3 Brooklyn startups featured in WeWork’s new WeMRKT
How Nava PBC helped government make Medicare’s largest change ever
How a Brooklyn CEO changed his life through meditation
Brooklyn’s Call9 wants to spend $24 million to keep people out of the hospital
Sign-up for daily news updates from Technical.ly Brooklyn