Swiggy builds generative AI-powered search for personalised recommendations

Food aggregator platform Swiggy is building generative AI-led solutions to streamline its food delivery, quick commerce and dineout businesses.

Dubbed neural search, Swiggy’s new AI features will enable users to search for dishes and restaurants using conversational and open-ended queries and receive recommendations tailored to their specific needs, the company said.

“The neural search capability has been built using a Large Language Model (LLM) adapted to understand dish, recipe, and restaurant terminology, as well as Swiggy-specific search data,” said Madhusudhan Rao, Chief Technology Officer at Swiggy, in a blogpost.

The service will begin as a pilot in September, after which it will be rolled out to all search traffic on the app based on the initial customer response.

“With over 50 million items in our food catalogue, we fine-tuned the model through a meticulous two-stage process to ensure accurate and real-time responses to relevant food-related queries,” Rao said.

The company also claims to have “enriched” its catalogue with images and detailed descriptions using generative AI techniques. Neural search will soon support voice-based queries and queries in select Indian languages as well.

The Bengaluru-based food aggregator currently has over 250,000 restaurant partners across the country.

The company is also integrating neural search into its quick commerce business, Swiggy Instamart, where customers will be able to discover groceries and household items in a conversational manner.

Likewise, the firm’s restaurant discovery platform, Swiggy Dineout, will also leverage generative AI techniques via its Dineout conversational bot.

The bot will act as a ‘virtual concierge’ and guide customers to restaurants that meet their preferences, be it ambience, kid-friendliness, valet parking, ratings, cost, etc., the company said.

“We are collaborating with a third-party to develop a GPT-4 powered chatbot. building generative AI-led solutions to better serve our restaurant and delivery partners. For example, we are piloting in-house tuned LLMs to empower restaurant partners to self-serve on processes and questions related to onboarding, ratings, payouts, etc, leading to faster issue resolution and streamlining,” said Rao.

A conversational assistant powered by this LLM will be available in the restaurant-owner app and via WhatsApp.