Ria 2.0: Harnessing Generative AI for a Healthy Diet

Recently, Indian health and fitness startup Healthify (formerly known as Healthifyme), supported by Khosla Ventures, launched Ria 2.0, a generative AI-driven personal health coach with multi-mode and multi-language conversation capabilities.

The health advice provided by this chatbot is adjusted based on individual lifestyle and goals, taking into account factors such as schedule, meals (captured through photos), blood sugar levels, activity, and dietary preferences.

Healthify introduced the "Snapshots" feature this year, which revolutionizes nutrition tracking by allowing users to take photos of their meals for instant identification of various food types and calculation of nutritional content (including calorie counting). The company claims it is more than 40% more accurate than the original version and can identify over one million global food items.

Two additional announcements were made at the event. First, the fintech company has entered into a commercial partnership with food commerce platform Swiggy, allowing users to order meals through the Healthify app based on Ria's intelligent dietary advice, making it easier to maintain a healthy diet.

Second, Healthify's "Coach Co-Pilot," launched in 2019, combines AI with human coaching methods and has improved weight loss effectiveness by 70% after validation by Stanford research.

Insights into Healthify's Generative AI

"Our approach involves using multiple base models to contextualize information based on specific tasks, rather than relying on a single model to solve all problems," co-founder and CEO Tushar Vashisht told AIM at the event.

Healthify's team uses solution models based on various statistical models, including AWS, Meta, OpenAI, and open-source options, focusing on domain knowledge to gain new insights.

"Leveraging an extensive dataset of over 350 million messages and meal plans, their approach involves assembling different models to target specific outputs, such as improving engagement in voice interactions analyzing user data," said Abhijit Khasnis, VP of Technology, to AIM. When standard base models cannot meet the requirements, the team strategically utilizes specific use cases to interact more effectively with real-world data.

Another important point is the emphasis on user control and confidentiality when handling health data. Customers can delete their data and decide whether to share it with their coaches. "We never engage in data collaboration or sell user data, which aligns with regulatory and ethical considerations. Customers have complete control over it," Vashisht pointed out.

Growing with AI over the Years

According to the company, a significant aspect of its journey is the transition from old AI systems (such as the original Ria launched in 2019) to the new generative AI system.

"Our initial AI could only handle a limited range of queries and data points. However, with technological advancements, we expanded the data vectors to include inputs such as calendars, geolocation, heart rate, nutrition, and fitness, greatly enhancing the comprehensiveness of our services," he added.

Bhojarajan agrees with Vashisht's viewpoint, recalling that when the team first considered monetizing its AI services in 2022, which was when ChatGPT was born, it was another turning point.

"Initially, we thought it would only improve conversations, but it quickly became apparent that this new mode of communication would improve the application interface, leading to the development of a new application with a chat messaging system, marking a shift from traditional input methods to a more unified and interactive interface," added Bhojarajan.

Currently, Healthify is preparing for international expansion, and despite last year's restructuring efforts and layoffs, this fiscal year is expected to be the most profitable year to date, according to Vashisht.

"For FY24, we expect double-digit growth, revenue exceeding $30 million, approaching a $40 million run rate, while maintaining unprecedented low costs," he concluded.