Two Indian Engineers on a Mission to Automate Home Cooking for the World

Building a robot that performs mechanical cooking actions is straightforward engineering. Creating one that thinks, perceives, and improvises like a human cook presents an entirely different challenge.
Is the tomato purée thick enough? Do the onions need a few more seconds of sautéing?
Posha, a San Francisco-based startup founded by two Indian engineers, Rohin Malhotra and Raghav Gupta, is pursuing this challenge.

In an interaction with AIM, co-founder and CTO Rohin Malhotra outlined how the company’s appliance transforms raw ingredients into ready meals.

For users overwhelmed by AI products that just write emails and generate images, Posha represents a different league: AI applied to physically demanding tasks that people may want to avoid.

The startup offers an early glimpse of AI handling domestic work that requires real-world perception and judgment.

An Attempt to Bridge the Gap

Posha’s hardware features mechanical arms that pour and stir ingredients, as well as dispense spices through multiple pods, along with an induction pan and integrated oil and water tanks. There is also a display through which users can interact with the appliance, view setup instructions, recipes, and more.

“The first step would be to choose how many people you’re cooking for,” said Malhotra, as that would guide Posha about the quantity of ingredients to be fed in. The appliance can churn out up to 600 pre-programmed recipes for up to four people.

Thanks to AI models equipped with computer vision hardware, Posha can ‘watch’ food change during cooking and make real-time decisions about when to adjust the heat, add more ingredients, or proceed to the next step.

Besides the recipe, Posha can also accommodate for any missing ingredients or specific dietary restrictions intelligently.

“This happens a lot when some of our customers don’t have enough time, or the lack of skills to prepare the ingredients the right way — Posha can detect that and ensure the final recipe turns out the same way,” said Malhotra.

The appliance is priced at a one-time fee of $1,499 and is being sold in the United States in limited quantities. The company recently raised $8 million in Series A funding led by Accel Ventures.

In a live demonstration for AIM, Posha prepared paneer tikka masala, a traditional Indian main course made with cottage cheese, in approximately 25 minutes.

Where’s the Training Data?

Essentially, the company needed to train the computer vision models to cook all 600 recipes in its database. “One of the challenges was that there was no data on which these models could have been trained. We had to create our own datasets,” said Malhotra.

The team had to break cooking into component skills that function like “Lego blocks” for recipes. For example, to teach the system frying, engineers cooked 10 different ingredients from raw to burnt, training the camera to recognise colour states from raw to golden brown to black.

Similarly, when cooking ingredients that shrink in size, such as mushrooms, the camera can calculate the percentage by which ingredients’ size decreases, said Malhotra.

“We had to get a large number of ingredients initially and then create a substantial amount of synthetic data. We also had to cook them and gather all the necessary data before training our models on that,” he added.

Each time a user cooks a recipe, Posha also collects camera vision data to enhance the models’ efficiency, Malhotra said.

To achieve the desired outcome, individuals from various fields of expertise were assembled. The company collaborates with mechanical, electronics, manufacturing, software, and AI engineers — and also, chefs. “A few people who work here were chefs in their previous jobs. And now, for their living, teach robots how to cook.”

Taste Over Tech

None of these efforts that went into development would matter if the food doesn’t taste good. The looks of the finished appliance and its capabilities would all seem worthless.

“One of the most important aspects which we discovered was that we need to get people to taste the food,” said Malhotra. “What users are really sceptical about is whether Posha can cook good food.”

“The easiest way to do that is to do a lot of demos. We invited people to our office, and a lot of them became our customers after seeing a demo and tasting the food,” added Malhotra.

The company’s design philosophy, he said, focuses on providing the user a clear view of how the appliance is cooking their food. “It helps build a level of trust.”

Furthermore, saving time or effort is not the only goal of Posha, but also promoting a healthy diet.

“The amount of time people spend cooking food at home [in the US] is extremely low. That is increasingly being replaced by processed food from supermarkets or food delivery apps,” said Malhotra. Posha’s library of recipes also contains a long list of health-based dishes across various cuisines and dietary preferences.

Having said that, Posha isn’t the only company in the smart cooking or cooking technology sector. Startups such as Tovala, June Oven, Anova Culinary, and Impulse Labs offer a variety of products and appliances in the smart kitchen market.

While most leading AI-enabled cooking solutions are offered by startups and modern companies, it would be interesting to see how mainstream appliance manufacturers respond to both the opportunity and the challenge.

The question that will drive innovation remains the same: how long will it take for incumbents to develop such solutions at scale, and at a fraction of the cost?

The post Two Indian Engineers on a Mission to Automate Home Cooking for the World appeared first on Analytics India Magazine.