Close Ad

The Spoon

Daily news and analysis about the food tech revolution

AI

This Startup Is Using AI to Bring “Post-Organic” Farming to the (Urban) Masses

Kale: great for your health, not so great for the tastebuds. Sometimes I wonder if people eat it because they actually like the taste or because it’s so trendy.

That issue may soon become irrelevant, however, thanks a company called Bowery, which is using artificial intelligence (AI) to tweak crops’ color, texture, and even taste.

Billing itself as “The Modern Farming Company,” the New Jersey-based indoor-farming startup will soon open a second facility it says will be the most technologically sophisticated in the world. Sounds like a brazen claim, until you look at what Bowery actually has cooking, er, farming, at its forthcoming facility.

Bowery’s “brains” are found in its propriety system called FarmOS. Using vision systems and machine learning, FarmOS monitors the crops 24/7, collecting data about water flow, light levels, temperature, and humidity. Bowery growers can then use the data to make adjustments to the environment, which will impact color, texture, and taste. The system also alerts growers when plants are ready for harvest.

All of those elements and more roll up into what Bowery founder Irving Fain recently called “post-organic produce”—Bowery commands the entire process of raising produce, from seed to store, and grows crops in a fully controlled environment that doesn’t have to rely on chemicals, pesticide, or human intuition to ensure quality of crops. Sure, the name’s a little much, but the concept grows more promising each year, thanks to factors like cheaper LED lighting, better data analytics, and concepts like vertical farming, which is predicted to be worth $13 billion by 2024.

And while they’re not all using the “post-organic” label, there are plenty of others exploring the possibilities of farming in fully controlled, indoor environments.

Also in New Jersey, AeroFarms has a 70,000-square-foot facility, where it grows bok choi, arugula, watercress, and other greens, including kale. The company closed a $40 million Series D funding round at the end of 2017, bringing in IKEA Group and Momofuku’s David Chang as additional backers.

Meanwhile, indoor farming startups abound in Alaska, where growing produce outside is pretty much impossible in the depths of winter and anything shipped is often close to spoiled upon delivery. Alaska Natural Organics operates a 5,000-square-foot farm that grows butter lettuce and basil. Vertical Harvest Hydroponics designs systems that can be grown inside shipping containers and distributed across the state, including hard-to-reach areas. Both companies are based in Anchorage.

And in Kyoto, Japan, a “vegetable factory” is run by robots and grows 30,000 heads of lettuce per day. The company, Spread, says that it recycles 98 percent of its water and, because the factory is sealed, doesn’t have to rely on pesticides or chemicals.

What sets Bowery somewhat apart—for now, at least—is that it has gone beyond simply monitoring water supply and temperature with its ability to adjust things like taste, texture, and even blemishes on produce. With the U.S. alone throwing out about 50 percent of produce grown annually, a proprietary system like Bowery’s could seriously be leading the way in terms of indoor farming’s impact on overall agriculture.

 

Knorr’s “Eat Your Feed” Delivers Instagram-Inspired Recipes… Sorta

Millennials everywhere can finally justify all those overhead photos they just had to snap (and then Instagram) before digging into their food.

Knorr, the powdered soup and seasoning brand owned by Unilever, has developed an AI-powered tool which scans your Instagram feed and then recommends recipes based on your photos. Dubbed Eat Your Feed, the tool uses visual recognition technology to match your food snaps with recipes from Knorr’s database. After you get your recommendations, you can save the recipes or add the ingredients to a digital shopping basket. And if you’re not already on the ‘gram, don’t worry — you can use this short quiz on Knorr’s website to get personalized recipes.

In the spirit of thorough journalism, I decided to give Eat Your Feed a try.

After entering in my Instagram login information, the webpage whirred around a bit before directing me to a page of almost completely nonsensical recipe matches.

First up was a photo I took of burgers & fries (it was actually the Impossible Burger, but I wouldn’t expect Eat Your Feed to know that). I would have expected it to match this to a perhaps another burger recipe, or even a grilling one, but instead I got… chicken and pasta soup?

As I scrolled through my recommendations, some of Eat Your Feed’s logic became clear. Some. For example, a photo I’d posted of some seaside cliffs linked to a recipe for Mussels Meuniere. However, most of the tool’s process was still shrouded in mystery: why did a painting of a cake equate to spinach soup? What linked a photo I took of a cave in Greece to a lemony pasta dish?

Presumably, it was some tag which I didn’t know existed — but Eat Your Feed did. When you allow the tool access to your Instagram, it also gains access to all of your data stored on the platform. It uses AI to scan your captions, locations, and tagged people to try to draw links to their recipe database. A few of these tags are displayed above your matches, which gives you a clue into how the algorithm made its selections. This explains why my photo of the ocean synched up to a mussels dish — both were tagged “Beach.” As to how they categorized both burgers and chicken & pasta soup as “Time” and “United States” is slightly less clear, however, though I suppose I was in America when I ate them?

Those who take the quiz instead of letting Instagram take the wheel have a bit more transparency into their recommendations. I took the 5-question quiz and was suggested recipes that were “Active” and “European” based on my answers. Which makes more sense than pairing cake and spinach soup together because both are purportedly “Swedish.”

One of the tool’s biggest issues lies with Knorr itself. All of the recipes must contain at least one of their ingredients, and since Knorr only makes soup stock cubes and powders, that limits the selection pretty severely.

I couldn’t find a way to save a recipe or add ingredients to an online shopping cart; the only option was to email myself a link to the recipe. In the future, it would be smart for Knorr to partner up with a shoppable recipes platform and a grocery delivery service like Allrecipes/AmazonFresh so they can actually deliver on those promises.

To give Eat Your Feed credit, the tool was gimmicky enough to suck me in in the first place. Plus, I did find myself clicking around other recipes on the site for a minute after I got my personalized meals. However, most of my suggested meals were so laughably off-base that I’m wouldn’t be inclined to make them at home, no matter how much they might remind me of that time I went to the beach two years ago.

So is the tool worth using? In short: no. It gives you almost no utility, but it’s still fun in the way that, say, taking on online quiz about Which Backstreet Boy Is Your Spirit Animal is fun: it’s pretty useless and probably inaccurate, but it’s a great way to waste a few minutes on the internet.

To promote Eat Your Feed, Knorr will open a pop-up restaurant at London’s Jones & Sons on April 11, where diners will be served meals matched to their Instagram feeds. I’ll be sad to miss my four-course meal of various soups and soup-like dishes, but maybe I’ll check in on the ‘gram. And then recreate them at all at home.

Farmstead Raises $2M, Releases FreshAI for Retail Food Management

It’s only Tuesday and already it has been a busy week for Farmstead, the startup that uses a combination of machine learning and small food hubs to enable home grocery delivery. On Monday, the company announced it had raised a $2 million seed round, and today, Farmstead opened up its FreshAI predictive analytics software platform to external food companies.

Farmstead is trying to create an entirely new model for grocery stores by limiting its physical footprint and using AI to intelligently curate its inventory. If successful, Farmstead Co-Founder and CEO, Pradeep Elankumaran, believes his company can drastically reduce the huge food waste problem in this country (where 40 percent of our food is never eaten).

Traditional grocery stores are massive buildings stocked to the gills to provide shoppers with a ton of options. Farmstead, on the other hand, uses food “microhubs” that are only about two to three thousand square feet and carry only about 1,000 items across all shopping categories.

It can get away with carrying such a small inventory because its AI platform predicts the items customers will want and orders just the right amount. To do this, Farmstead combines historical sales data and current trend data with consumer recommendations and external factors, such as holidays and product sell-by dates. Its machine learning processes these data points to assess which items to carry and how many it should stock at any given time.

“The system gives us a very fine prediction for stock,” Farmstead Co-Founder and CEO, Pradeep Elankumaran told me. “We want to have just enough on the shelves so no new customers walk away disappointed, but not so much that product gets wasted.”

Because Farmstead microhubs are small, they can be implemented closer to residential neighborhoods rather than warehouse districts to facilitate faster home delivery. Farmstead currently operates in the Bay Area and offers one-hour delivery from 8:00 a.m. to 9:30 p.m. for people in San Francisco, as well as rolling delivery hours during the morning, afternoon and evening for those in the elsewhere in the Bay. You can also drive to the Farmstead hub in San Francisco or San Mateo and a worker will run out and put your groceries in the car.

Elankumaran said that Farmstead is now making 1,500 – 1,600 deliveries per week. With this $2 million round, Farmstead has raised a total of $4.8 million so far. It will use the new money to scale its delivery service and expand beyond the Bay Area and will build out its core AI technology.

Which brings us back to today’s news about FreshAI. Farmstead is opening up its predictive analytics platform to supermarkets, cafeterias, restaurant chains and other companies that work with food at scale, allowing them to use Farmstead’s automated inventory management system. Farmstead says that using FreshAI has reduced its perishable food waste to sub ten percent as opposed to “35 – 40 percent average perishable food waste for typical supermarkets.”

For now, FreshAI is a pilot program, and companies can apply to take part on Farmstead’s site. If accepted, companies will upload operations data, and Fresh AI will provide daily and weekly order recommendations.

Farmstead is part of a new generation of startups such as Spoiler Alert, Ovie, and LeanPath that are using technology to combat food waste. It’s also reminiscent of CommonSense Robotics, which is using a combination of robots and micro-fulfillment centers. But Elankumaran said Farmstead is still a ways from using robots in their hubs or to fulfill deliveries. Instead, he said that they’re focusing on getting the best possible customer experience.

If Farmstead can deliver on that experience as it scales out, the busy weeks won’t be ending anytime soon for Elankumaran or his company.

Services that Combine Flavor and AI Are a New Food Tech Trend

Artificial Intelligence is making its way into our food system in a big way. It’s on dairy farms monitoring milk quality, in restaurants powering food-running and burger-flipping robots, and even in the kitchen, walking you through a recipe in the guise of a voice assistant or chatbot.

Lately, we’ve noticed AI playing another role in what we eat: this time in flavor development. We’ve rounded up 5 startups merging AI and flavor to help restaurants and consumers create more sophisticated dishes, teach home cooks how to make dinner, and reduce friction for food R&D.

Foodpairing

Foodpairing is a platform which uses machine learning and data analysis to create a sensory map detailing which foods taste good together. Since roughly 80% of taste actually comes from smell, they base their findings on the aromas of each ingredient. The Foodpairing Inspire Tool allows their customers—mostly professional chefs and bartenders looking to create innovative, unexpected dishes no one has tasted before, but also home cooks—to discover pairings of the more than 2,500 ingredients in their database. It markets itself as having pretty wide applications, powering everything from smart kitchen apps, e-grocery, personalized recipe and drink recommenders, and mHealth.

PlantJammer

This app (which is currently available exclusively on their website) grew out of an ex-consultant’s desire to teach himself how to improvise in the kitchen. Using flavor mapping technology similar to Foodpairing’s—both are based around aromas and use machine learning—the platform allows users to select complimentary ingredients based on what they have in their kitchen. Once the selection is complete, the algorithm generates a custom recipe. The Copenhagen-based startup hopes to use their AI-driven platform to promote plant-based cooking and reduce food waste.

dishq

Self-described “food AI company” dishq uses customer data, machine learning, and food science research to predict consumer taste preferences. They translate their findings into APIs to help their clients, which range from food delivery platforms to corporate cafeterias, provide tailored food suggestions to their customers and outline emerging food trends. As co-founder Kishan Vasani told the Spoon, dishq offers “taste analytics as a service,” allowing companies to react quickly to food trends as they are happening.

FlavorWiki at work quantifying data to report on top food trends.

FlavorWiki

FlavorWiki uses analytics to measure consumer taste and dietary preferences. They aim to unlock new applications for taste data across the food system. While they market themselves to a wide audience—everyone from major food companies to moms with picky kids—their taste-profiling technology is chiefly aimed at retailers. By creating self-described “taste archetypes,” FoodWiki hopes to help clients like CPG companies cut down on R&D costs for new products, reducing the pricey trial and error stage. If you’re curious about how exactly the FlavorWiki system works—and where it hopes to go—give our podcast with their CEO and Head of Product Daniel Proz a listen.

Gastrograph

Gastrograph is another company using AI to help food & beverage producers streamline new product development. Their technology maps the flavor preferences of individual consumers and also predicts broader consumer reception to new taste profiles. Gastrograph hopes to help create only slam-dunk food products by using machine learning and predictive algorithms—no more costly duds. If you want to hear more about this AI-driven food tech company, check out our podcast with Gastrograph CEO Jason Cohen.

For food startups and CPG developers struggling to differentiate themselves from their competitors, services that use AI to predict and develop delicious, memorable foods would be a useful investment. If flavor/AI companies can deliver on their promises—to cut R&D costs, to help chefs and home cooks create tasty recipes, and to predict emerging food trends—they could be that extra something that spells success for emerging companies, or for big food giants whose current products are starting to feel stale. Flavor/AI technology could also play a huge role in predictive restaurant ordering or grocery delivery, both of which Amazon has in the pipeline.

The bottom line for food industry folks, if you don’t have a taste for AI, you’d better develop one—and soon.

P.S. The CEO’s of Dishq and Foodpairing will be speaking at SKS Europe in June! Register today to hear them talk about how AI will change the way we buy & eat food. 

Check, Please! Ex-Googlers Bring Robot Food Runners to Restaurants

Are you ready for even more robots in your daily life? Yes? Good—because they might soon be whirring along through your favorite local pizza joint, delivering your pepperoni pie, clearing up the scraps, and bringing you the check.

Silicon Valley-based startup Bear Robotics has created robots to bus tables and deliver food in restaurants. Their AI-driven robots—which vaguely resemble hip-height bowling pins with a flat disc on top—use self-driving technology to navigate through restaurants, avoiding people and other obstacles. 

When food is ready to run, managers simply use a tablet with Bear Robotics’ connected app to call a robot (dubbed “Penny”) to come pick it up and whisk it off to hungry customers. When they’re finished eating, Penny will return, ready to deliver dirty dishes to the kitchen before bringing the bill. 

If you’re thinking that this sounds very Silicon Valley-esque, you’re not wrong. In fact, Bear Robotics co-founder and CEO John Ha spent six years as an engineer at Google before turning his sights on the hospitality industry. He invested in a Korean restaurant, which is where he started to realize how inefficient and difficult foodservice could be.

“[Servers] are tired, they get a low salary, usually no health insurance, but they’re working really hard,” Ha told The Spoon. His company wanted to streamline the point of sale and increase efficiency in front of house operations. So they decided to build a runner robot who could deliver food quickly and clear dishes. 

Hence, Penny was born. Well, made.

The company currently has one robot deployed at Ha’s restaurant Kang Nam Tofu House in Milpitas, CA. When it was first introduced in August 2017, engineers had to be on call to make sure that the robot was running food smoothly. Since December, however, Ha says that the robot has been deployed every day, unsupervised by engineers. It now works seven days a week.

The robots are controlled by an app which Ha claims is very user-friendly. Restaurant owners or managers can simply mark key locations (Table 1, kitchen, etc.) and the robot will learn the layout by moving around the space, drawing a map as it goes. When food is ready in the kitchen managers use the app to call a Penny, who then runs the dishes to their matching customers. 

Penny doesn’t have arms, though, so either servers or the customers themselves still have to transfer the food from the robot to the table. This means that the robots don’t necessarily save a lot of time, though Ha claims that that time adds up. By not spending time waiting in the kitchen for food to be ready or running back and forth to the kitchen with dishes, servers are able to stay in the dining room and chat with diners.

Ha is in the process of selling his flagship restaurant to focus on expanding Bear Robotic’s robot food-running empire. The company plans on renting out Pennys using a labor as a service business model to local chains for an hourly or monthly fee. So far, local pizza chain Amici’s Pizza has signed on to add food running robots to their staff once a week. Ha is also in the process of opening a Japanese ramen place which will “employ” three robots to serve and clear dishes. 

So how do customers feel about having their entrées delivered by a robot that sort of resembles a slimmer R2D2? According to Ha, most of them are all for it. “People love to interact with the robot,” he said. “They especially love paying the bill when the robot brings it to them.” 

In the future, BearRobotics is hoping to expand Penny’s functions beyond simply running food back and forth. Ha envisions a future when the AI system can integrate more fully with POS, possibly even taking orders directly with customers without the guidance of an app.

Bear Robotics PennyBot demo

Unlike other companies who are applying AI to food production with burger-flipping and salad-mixing robots, BearRobotics is focused solely on the front of house operations. “By focusing on the front of house we [can] have a better impact on the industry and reach a bigger market with a simple, versatile product,” said Ha.

In other words, not every restaurant needs a robot with a specialized preparation skill like crepe-making, but most of them do need someone (or something) to run food and bus tables. That means that Bear Robotics’ Penny robots could function in virtually any type of restaurant, regardless of layout or type of cuisine.

I was surprised to learn that Ha doesn’t envision his robots (or their future iterations) completely replacing human servers. Instead, his goal is for their robots to help bulk up staff during busy times and decrease labor intensity on employees. He hopes that Penny can shoulder the physically draining parts of service, like dish clearing and delivering heavy plates, so waiters can focus on the more fulfilling aspects of their job, such as chatting with customers.

Integrating robots into the staff can also help human servers make more money—at least in theory. If a robot is covering food running and bussing, restaurant managers can schedule fewer servers per shift. That means that servers don’t have to tip out their bussers and runners at the end of the day. However, if a robot is delivering customers’ food, they might not feel the need to tip as much as they would with a human server—no matter how good of a job it does. 

Despite Ha’s assurances otherwise, I can’t help thinking that Penny (and other restaurant robots) still bring us one step closer towards an automated restaurant experience. We’re already seeing automated restaurant experiences in fast-casual joints like Eatsa and coffee shops like CafeX, but Bear Robotics is hoping to incorporate robots into nearly every type of eating establishment.

This isn’t necessarily a bad thing: front of house automation could well lead to more efficient ordering and food delivery, so you’ll never have to deal with cold fries again. But it is worth noting that Penny does cut out two full-time restaurant jobs which are pretty commonplace at high-volume and fancy establishments: bussers and runners. No matter how much it frees up servers to spend more time asking about your day, it still probably means fewer hospitality jobs for humans. 

The effectiveness of server robots will most likely depend on the type of restaurant experience. At a low-key pizza place I most likely don’t need any help ordering (anchovies, of course), but for a more fine dining experience I appreciate a more, well, human touch. Servers guiding me through the menu, explaining specials, and recommending dishes in an engaging manner is part of what makes forking out for a big meal worth it.  

Ha told us that he wanted to create a company that was “the Google of the restaurant field,” and in a way I think he’ll succeed. I could envision a future where Pennys are pretty ubiquitous, especially as their self-driving technology inevitably improves. Plus, if Ha’s predictions are true, Bear Robotics will expand their robots’ operations until they could effectively run the entire front of house by themselves. So if you weren’t already mentally prepared for robots to enter every aspect of your life—even your pizza parlor—now’s the time. 

Goodbye Hanger Gap: Amazon Awarded Patent For Predictive Restaurant Ordering

You know the time between when you order a meal for delivery or takeout and when it arrives? The ‘hanger gap’? Well, Amazon wants to eliminate that waiting time between when you get hungry and when the food arrives by predicting exactly when you want to eat and preemptively ordering a meal for you.

How exactly would they do that? According to a patent just issued to Amazon on February 20th for “Predictive Restaurant Ordering”, they would use an AI-based modeling engine to anticipate customer needs based on variety of inputs. These could include past orders, as well as contextual information such as calendar appointments, location, caloric intake for the day and the amount of exercise a person has had. The patent suggests that one way it could access this information is by obtaining it from personal electronic devices such as smartphones, wearables, and personal computers.

Predictive Restaurant Ordering Flow Chart, from Amazon Technologies

The proposed system could also take a stab at what restaurant a person would want to order from that night by analyzing past behavior and using inputs such as various check-ins they’ve had on social networks like Facebook.

It’s fascinating – if not entirely surprising – to see the company working on anticipatory based ordering systems. I imagine they are probably working on similar anticipatory ordering for groceries, especially as the company doubles down on food retail with Amazon Go and Whole Foods.

And if all this sounds a bit creepy, that’s because it is. Tracking our behavior and shopping for us before we actually have the thought is just a bit too Black Mirror-ish for my taste, even if it would help me to avoid the hanger-gap.

Of course, it should be noted that Amazon hasn’t launched any service based on this technology, at least not yet. The company is amassing patents at a breathtaking rate, and there’s always a good chance they won’t ever launch services based on many of them. Conversely, Amazon is constantly pushing the boundaries around commerce, so I wouldn’t be surprised if we do see some form of anticipatory restaurant ordering someday from the Seattle tech giant.

In that case, for future reference, I’ll take a BLT on sourdough. But then, they probably already know that.

You can hear about Amazon’s ordering app in our daily spoon podcast.  You can also subscribe in Apple podcasts or through our Amazon Alexa skill. 

PlantJammer Uses AI To Create Instant, Flavor-Mapped Recipes For Home Cooks

What are you going to have for dinner tonight? Maybe a big bowl of cheesy pasta (me), a reheated plate of leftovers, or, for the more ambitious, sous vide steak? Some people don’t need to put any pre-planning into their evening meals; they can just throw together whatever’s lingering in their pantry and crisper drawers, improvising with what’s on hand. For others who aren’t comfortable riffing in the kitchen or who don’t have time to grocery shop for a particular recipe, dinner is often something requiring little-to-no effort and decision making. That can mean meal delivery kits with pre-portioned ingredients, or, more likely, takeout.

Vegetarian recipe-generating app PlantJammer is out to help those with low kitchen confidence who want to cook healthy meals and reduce their food waste. The app creates custom recipes for users based off of whatever ingredients they have in their kitchen—then walks them through how to go from recipe to meal, step by step.

The app is able to do all of this thanks to AI, which maps out ingredients’ elements based on their aromas, creating a sort of flavor fingerprint. They then use the aromatic profiles to draw links between seemingly disparate ingredients, suggesting to the user which foods would go well together. In this way, PlantJammer hopes to gamify cooking with plant-based foods, making vegetarian cooking less of a chore and more of a convenient, efficient way to create a meal.

PlantJammer isn’t the only app using AI technology to suggest new flavor combinations. There’s Foodpairing, a tool which also finds and analyzes compatibility between different ingredients, which Haase turned to during his initial forays into cooking. However, while Foodpairing seems to aim its services at food industry professionals looking to create innovative and unexpected dishes, PlantJammer is a tool intended to help home cooks find their sea—er, kitchen—legs.

In fact, PlantJammer originally came about because the founder, Michael Haase, needed help throwing together plant-based meals for himself. Before founding the Copenhagen-based company in 2016, Haase worked consulting on sustainability and resource management at McKinsey and Danish biotech company Novasymes.

A few years ago Haase decided to work towards making his eating habits more sustainable by doing two things: stop wasting food, and eat less meat. He wanted to learn how to improvise in the kitchen, making use of any lingering produce before it went south—but he also didn’t want to spend 10,000 hours learning how to intuitively cobble together a delicious meal.

So what does an ex-consultant do? First, they collect data—lots of it.

“I decided to bootstrap that learning, so I turned to my background in econometrics,” Haase said. He took the neural network model, the workhorse of AI, and applied it to cooking. “I collected the intelligence of thousands of years of humans learning to cook and used that as a data set to create patterns and, ultimately, build a landscape of taste.” This analytical tool can look at big data and find patterns to determine which aromas—and thus, which flavors—will work well together.

As Haase describes, it, the neural network is a sort of color wheel for taste. At the center of the wheel is salt. On top of that the app must balance four components that, at least according to Haase, every good recipe needs: acidity, umami, crunchiness, and mouthfeel (oil). You can add balancing touches on top, like spiciness, too. This technology can lead to some surprising pair-ups. For example, Haase claims that bananas and zucchini are a match made in heaven—one I have yet to sample.

As of now, PlantJammer has a neural network of 3 million recipes and 1000 ingredients.

While the PlantJammer model gets really granular (mapping all 148 aromas in asparagus), they also generalize—quite a lot, in fact. “We say that, at the core, there are only 9 recipes in the world, and then there are infinite variations on those recipes which we can modularize,” said Haase. Judging from the PlantJammer app, these recipes include quiche, salad, pasta, and soup—a list that, as expected, generates some pushback for both what is included and what it doesn’t. But Haase isn’t one to adhere to tradition, especially in the kitchen. “Who says you can’t put curry in the risotto? That’s one learning of management consulting: just because people have been doing something one way, doesn’t mean it’s the only way to do it—or even the best way.”

I decided to put PlantJammer through the test and take a spin through its app (currently available only through their website).

A prototype of the PlantJammer app.

When you first open the app, you are met with a selection of suggested recipe templates ranging from Roasted & Toasted Soup to Asian Quiche to A Freestyle Pasta. If those templates aren’t for you, you can create your own recipe and just “Jam.” Never one to be pinned down, I decided to freestyle and was led to a new page by a tiny eggplant in shades playing the saxophone (his name is Eddie). From there, the app prompts you to select 1 to 3 ingredients from each of 4 categories: bulk (vegetables and plant proteins), splash (vinegar, citrus juice and oil), boost (chilis and aromatics), and topping (herbs, nuts, and other garnishes).

I selected chickpeas and broccoli from the bulk category, and the other columns immediately rearranged themselves, placing the AI-generated best pairings for my selections at the top. I selected tahini, harissa, and sunflower seeds, then threw in some yogurt for good measure. After I’d made my choices, I was led to a customized 6-step recipe that told me how to transform my selected ingredients into a finished dish: Chickpea Salad. The name itself was somewhat bland, but I was impressed with how detailed the recipe was; it gave clear cooking times for each ingredient and made each step seem simple yet doable. More importantly, it sounded like the end result would taste good. 

PlantJammer still has room for improvement, though, if it’s aiming for mainstream acceptance—especially within an American audience. Some of their ingredients are confusing to decipher (“soy bean sauce” and “artichoke hearths”), and then there’s the fact that users are limited by the ingredients options given. What if I have a can of lima beans, which isn’t on the PlantJammer list, but no chickpeas, which are? An experienced cook would know to go ahead and substitute one for the other, but the app is geared towards a more novice audience, who might not feel as comfortable with ingredient riffing.

Kinks in the system aren’t the only hurdles that PlantJammer is facing. Haase admitted that some potential angel investors decided to pass on the startup because the app purposefully doesn’t include meat in its ingredient list. And they never will. For Haase and his team the choice to bypass meat is crucial to the company mission to promote sustainable eating habits.

And they might have gotten lucky with their timing. Plant-based proteins are having a moment, racking up funding and huge social followings. While PlantJammer situates itself as separate from the processed, lab-made meat and meat alternatives of Silicon Valley, if it succeeds, it will probably be in part thanks to their efforts. By making plant-based diets and cultured meat not only acceptable but admirable and—dare we say it—cool, companies like Memphis Meats and Impossible Foods are paving the way for other startups in the meat-alternatives sphere. Though it’s an app, not a product, PlantJammer can only succeed if it has a hefty client base willing to eat vegetarian meals for at least for a few nights a month.

PlantJammer isn’t the only app bringing modular cooking—or cooking with dynamic recipe templates—to consumers. Connected cooking platform Innit (which celebrity chef Tyler Florence spoke about at last year’s Smart Kitchen Summit) recently launched an app similarly creates recipes built on whatever users have in the fridge. However, while PlantJammer starts from scratch and shifts its suggested ingredients based on consumer inputs, Innit uses recipe templates which users can customize and tweak. It seems modular cooking is a trend we’ll be seeing more of. In today’s world of customization and AI leveraging in the kitchen, it might be the way we’re moving.

“We want to make cooking convenient, not a compromise. That way, we can hopefully make a lot of people change their habits,” said Haase. Banana and zucchini stir-fry it is, then.

Wine-Searcher Builds Casey The Chatbot To Reach ‘Everyday Wine Drinker’

Since the time Wine-Searcher was founded by London wine merchant Martin Brown in the late nineties, the site has become one of the Internet’s go-to destinations to discover new wine. Over the past 18 years, the wine search engine has made a name for itself by pairing an extensive database of wines with the opinions of renowned wine experts like Jancis Robinson to help thirsty users find their next great bottle.

But if you just need to pick up a bottle for dinner tonight at the local wine shop or grocery store, you may not have time to sift through the millions of listings (about 9 million at last count) on the Wine-Searcher website or on their mobile app to find one. But that’s probably ok with Wine-Searcher, since nowadays they might just suggest you ask Casey.

Meet Casey The Chatbot

Casey is Wine-Searcher’s new wine chatbot. The bot, currently available in beta on the Wine-Searcher website or through Facebook Messenger, is a big strategic initiative for the company who sees Casey as a way to expand their addressable market.

“For us, (Casey) is moving us into the everyday wine drinker market beyond the wine expert,” said company spokesperson Suzanne Kendrick in a phone interview with The Spoon.

Kendrick explained the typical Wine-Searcher user ranges from wine enthusiasts who know enough to discern they “like New Zealand Pinot” all the way up to wine experts. However, the company feels there is a large swath of wine drinkers who just want a good bottle of wine and don’t have time to learn the difference between New Zealand and California Pinot.

Those drinkers just “want a recommendation, want a great wine at a great price, and they want to get it now and not wait for it to ship next week,” said Kendrick.

It’s for this consumer – the ‘everyday wine drinker’ – that Wine-Searcher built Casey.

Minimal Viable Product

The company has been working on Casey for about a year and has eight people working on the project according to Kendrick. To help them build the bot, the team has been working closely with Microsoft. That’s because the framework powering the bot’s ability to carry on a natural language conversation is Microsoft’s LUIS (Language Understanding Intelligent Service) platform.

According to Kendrick, the Casey is getting better and better and having conversations about wine, but she says the chatbot is still in the “minimal viable product” phase of development. While Casey is good at wine recommendations, it’s still learning to how to make food recommendations.

Casey, Wine-Searcher’s chatbot

I gave Casey a whirl, and it worked better on the Wine-Searcher website than in Messenger, where the bot told me the server was unavailable after I asked it for a $75 bottle recommendation to go with a hypothetical meal of fried chicken. On the company site, Casey asked me my wine preference (red, white, etc.) and price range and was able to recommend a bottle. When I chose a bottle of wine, it handed me off to Total Wine & More’s website for me to choose in-store or delivery.

This last part is important because Wine-Searcher makes much of its revenue through its relationships with large wine retailers like Total Wine & More. Other wine destination sites like the fast-growing Vivino, which just nabbed $25 million in funding, are taking more of a one-stop shop approach for wine buying by serving up recommendations and handing the commerce and delivery as well (it also looks like Vivino is working on its own effort to take its wine scanner capability from the app and put it into bot form).

For its part, Wine-Searcher is happy to stay out of fulfillment and shipping and just be the Internet’s wine experts helping the widest possible audience. With its chatbot, which the company is just starting to talk about, they hope to expand their audience and help take the company into its next two decades.

SomaDetect Uses AI to Help Dairy Farmers Improve their Milk

There is a milk glut in the U.S.. Technology has allowed dairy farmers to produce more milk than ever, but all this abundance has caused milk prices to plummet. It’s getting so bad that some farmers face selling off their cows.

While technology helped create this crisis, perhaps SomaDetect‘s technology can help struggling dairy farmers get out of it.

SomaDetect uses a combination of optical sensors and machine learning to help dairy farmers analyze the milk each cow produces to determine its quality. As SomaDetect CEO Bethany Deshpande explained it to me, farmers attach a small sensor box to the milking hose, which shines a light through the milk as it flows. Based on the scatter pattern of that light coming through, SomaDetect’s software can analyze what’s in the milk.

The company measures fat and protein as well the reproductive status of a cow and any residual antibiotics. More importantly, SomaDetect can look at somatic cells to detect Mastitis, a serious bacterial infection of the udder that is the most common disease among dairy cows and the number one cause of their early death. By quickly identifying cows with high somatic cell counts, farmers can better target treatment and to help prevent the spread of Mastitis.

Additionally, one cow can throw off somatic counts for an entire batch of product. By removing high-somatic cows, the farmer can lower the overall somatic counts of their milk and earn more money: “Farmers with low somatic cell counts get a bonus from the milk processor,” said Deshpande.

This type of deep inspection of milk has only recently become possible. According to Deshpande, the optical technology has “been around for a hundred years,” but advances in computer vision and machine learning means SomaDetect’s software can analyze vast sums of information in ways that were not possible even five years ago.

SomaDetect isn’t the only company using light to help dairy farmers. EIO Diagnostics uses multispectral imaging for Somatic cell counting, and Consumer Physics is putting its handheld SCiO device to use on farms to detect levels of dry matter in cow feed, which can also impact milk production.

SomaDetect was founded in 2016 in New Brunswick, Canada. The company won a 43North startup competition, earning them $1 million in funding and office space in Buffalo, NY. SomaDetect makes money by selling the equipment to farmers and charging $5 per month per cow for the software. The company is currently running pilot programs and is searching for seed funding as it looks to expand into New York state.

Deshpande says she came from the dairy side of the industry and wants SomaDetect to work closely with farmers. With milk prices expected to stay low throughout this year, dairy farmers could use all the help they can get.

You can hear about SomaDetect in our daily spoon podcast.  You can also subscribe in Apple podcasts or through our Amazon Alexa skill. 

Dishq Uses Machine Learning for Bespoke Food Recommendations

This is embarrassing to admit, but whenever I go to a Thai restaurant, I just order tofu phad thai. Always. Yes, that is totally generic, but I don’t get to eat out that often, so I don’t want to take a chance on something that I might really dislike. I know what I’m getting with the phad thai, so I settle.

It’s that settling for the same-ole that Dishq is looking to improve by using artificial intelligence. Based in Bangalore, India, Dishq provides APIs for food service companies like restaurants, corporate cafeterias or food delivery services, so those companies can implement AI-powered, customized food suggestions for their customers.

During an interview, Dishq Co-Founder and CEO Kishan Vasani told me there are four parts of his company’s offering:

  • A database of more than 100,000 dishes that are broken down into 26 different attributes including ingredients, cuisine style and cuisine origin
  • Anonymized customer behavior analytics data
  • A machine learning algorithm
  • Food science research from around the world that feeds into Dishq’s algorithm

Vasani said that the collaborative filtering used by Amazon or Netflix to make suggestions won’t work for food because meals are such a personal experience. “Too many things go into it,” said Vasani, “What you like, where you grew up, who you’re with.” He says that Dishq’s deep data-driven approach allows for truly bespoke recommendations because it understands food at the flavor compound level as well as transaction history.

Where the customer encounters those recommendations are up to the company using Dishq’s API. It can be used at the menu level to surface suggestions or email notifications. It depends on whether that client is looking for more conversions, increased average order value, or just creating a better customer experience.

Right now Vasani says that Dishq has 6 clients with two million recommendations generated every month, and that clients see an 11 percent uplift in revenue with Dishq.

Sadly, Dishq can’t offer a universal taste passport that travels from restaurant to restaurant. So what you like at Domino’s would help them determine what you might like at Dunkin’ Donuts. The reason for that, Vasani says is data protection and restaurants not wanting to play nice with one another.

Founded in 2015, Dishq has 13 employees and has raised $160,000 to date, with $120,000 of that coming from the Hong Kong-based AI accelerator Zeroth.ai. Vasani says he plans to start looking for new funds in Q2 of this year.

Vasani is also looking to expand beyond food service clientele and into the consumer packaged goods. He refers to the forthcoming product as “Taste analytics as a service,” and would allow CPG companies to react more quickly to food trends as they are happening. For example, if avocados were suddenly appearing everywhere on Instagram and social media mentions around Nashville, Dishq’s data could help a CPG company spot and understand those trends to quickly ramp up some avocado-related product for that location.

Until then, however, Dishq just wants to make eating out more pleasurable. Vasani wants Dishq’s recommendations to “shift people’s experience from a 6 out of 10 to an 8 or 9,” said Vasani. If Dishq works as promised, that could mean a lot less phad thai for me in the future.

Chefling Raises $1 Million for its Inventory and Recipe Assistant

Chefling, a kitchen app that connects what food you have with recipes and shopping lists, has raised $1 million in funding. According to a report in VentureBeat, the money will be used to for to hire marketing people, data scientists and a chef.

With the Chefling app, users can scan barcodes or take a picture of their receipt to monitor what foods they have in their fridge and pantry (an keep tabs on when that food will expire). Based on your food inventory, Chefling’s smart cookbook will then recommend recipes you can make. If you are missing any ingredients, Chefling automatically creates a shopping list for you and as you check these items off this list, the app keeps track of the new food available for newer recipe recommendations. Chefling works with Alexa and Google Home so users can ask for recipes or add grocery items to the list just by talking.

Chefling was created by a group of Northwestern grad students, and they were part of our Smart Kitchen Summit Startup Showcase last October. You can see their demo pitch on how Chefling works in this video:

Smart Kitchen Summit Startup Showcase Pitch: Chefling from The Spoon on Vimeo.

The nexus of food on hand, recipes and shopping is fast becoming a hot market with a number of entrants. LG and Samsung showed off smart refrigerators at this past CES that let you keep track of food and offer up recipes. Recipe app Innit lets you alter recipes based on different ingredients you already have. Smart tag Ovie keeps track of when your food will go bad and makes meal recommendations. Additionally, Flexy and AllRecipes are making shoppable recipes a reality by integrating with Amazon.

It’s also interesting that Chefling is hiring an actual Chef. In the VentureBeat report, the company says the chef “will be hired to curate and improve upon tens of thousands of recipes Chefling draws from food bloggers.” This perhaps points to the limitations of relying solely on algorithms and scraping websites for recipe recommendations. Or, just as Innit hired celebrity chef Tyler Florence, maybe Chefling will bring on their own celebrity chef to boost its visibility in a crowded space.

Chefling’s investment was from Chicago-based XVVC LLC, and brings the total amount raised by the company to $1.2 million.

‘Humanless Retail’ On Display at CES, But Will Humans Buy It?

One trend on display at this year’s CES is what I would describe ‘humanless retail’, where technology is used to sell physical goods to consumers without the help of humans.

Of course, this trend isn’t new. 2017 brought us a bunch of new ideas for taking the human out of the retail transaction by using machine vision/AI, IoT and more. What I saw on the show floor in Vegas is just a continuation of these concepts.

For example, last year we hear a lot about Amazon Go, a store concept where customers walk in and out without ever talking to a cashier. And this week, we saw the startup version of this in AIPoly, a company which offers a machine vision and sensor platform to create what the company calls “autonomous markets”.

Just as with Amazon Go, AIPoly customers register with the “store” and are identified as they walk in (or up to in the case of a kiosk) through facial recognition. The store then registers a purchase as the machine vision recognizes the products they pick off the shelves.

Below is a pic of the demo the company was showing off at CES.

And then there’s the Qvie, a single-product micro-vending machine that is essentially a connected lockbox version of the booze fridge in the Hilton. Qvie is targeted at the Airbnb host as a way to enable additional revenue through in-room sales, a trend that seems almost inevitable as Airbnb becomes a more and more viable alternative to hotel stays.

Finally, there’s Robomart, which can best be described as the love child of the controversial Bodega and an autonomous automobile.  The vision behind Robomart is a retailer such as 7-11 or Target would lease a fleet of Robomarts, stock them, and then bring the store to the consumer’s home. While it’s not exactly the same as Zume Pizza delivery trucks, it does something similar in making the retail location less relevant by bringing the point of presence closer to the consumer.

Robomart CEO Ali Ahmed told me he expects the first Robomarts to be available this year, which strikes me as extremely ambitious since the company is still raising funding to build out its vision. A mobile autonomous car-store combo doesn’t strike me as something you can do cheaply.

These are just three ideas I ran across in a couple hours on the floor at CES, enough to make clear that humanless retail is going to be much in 2018. The question for me is, will humans buy the idea of humanless retail, or is this just another case of Silicon Valley getting ahead of itself as it looks for addressable markets to apply new tech like AI, robotics and IoT?

The answer is yes, humanless retail is going to big. Sure, there will be lots of companies floating in the humanless retail startup deadpool before it’s all said and done (this is the case with pretty much every startup market in case you haven’t noticed), but the reason I think many of these early ideas will become much bigger and common is they’re simply evolutionary steps of what we’ve been seeing for decades and with much more rudimentary technology.

The self-service checkout at the grocery store, vending machines in your office, and the booze fridge in your hotel room are all innovations aimed at selling things to people without the need for another person to take money and put something in a bag. The only difference with these new ideas is the latest technologies to make humanless retail more convenient than ever before.