Places Elevates Dining Discovery Leveraging Amazon Bedrock to Achieve a 50% Boost in Arabic Search Performance

Digital startup Places aims to disrupt the hospitality market one major city at a time. The company, founded by Matthew Davidson and Walid Abou Ali, has created one of the first apps to digitally catalog menus throughout an entire urban center. Called Places UAE, the app provides access to over 3.5 million searchable food and drink options from more than 10,000 restaurants and cafes in Dubai. Davidson and Abou Ali felt existing dining discovery apps underserved Dubai’s food enthusiasts. Search functionality didn’t consider the city’s cultural mix and the variety of Arabic dialects used by residents. The founders also believed a more user-friendly way to discover the ideal place to eat was needed for local people with dietary restrictions and the 10 percent of Dubai citizens with food allergies or intolerances.

When the founders considered how to deliver their vision, they decided to put artificial intelligence (AI) at the heart of Places UAE’s search capability. People could then find their ideal place using conversational-style terms in Arabic dialects, with the technology yielding finely tuned results aligned with the users’ desired food, dietary needs, price, location, parking, ambiance, and more. After building a search engine and AI capability to support Arabic search functionality, Places looked to add a generative AI layer to the app’s existing platform. Abou Ali, cofounder and chief technology officer at Places, explains, “With our in-house search model processing data in English, we wanted to use generative AI for translations so it can understand different Arabic dialects.”

During initial experiments with generative AI, Places tested a standalone, publicly available, general-purpose translation model. However, Abou Ali and the Places team soon noticed several areas that needed improvement. Places found the model produced different translations when the same search term was given multiple times. What’s more, because communication occurred across the public internet, there were also concerns about a possible security risk. Plus, there was also unexpected downtime to contend with.

After a conversation with the local AWS account team, Abou Ali met with an AWS generative AI expert to discuss Amazon Bedrock, a fully managed generative AI service offering a choice of high-performing models for tasks, including translations. Based on the discussions, Places accessed $5,000 in AWS credits for launch tests on the generative AI service using different models for translations. “An important advantage of the AWS service was that we could seamlessly integrate the service with our cloud infrastructure,” states Abou Ali. Moreover, with Amazon Bedrock operating over an Amazon Virtual Private Cloud (Amazon VPC), Places alleviated security and data privacy concerns.

Places tested multiple large language models (LLMs) on Amazon Bedrock before adopting Claude 3.5 Sonnet. “During the tests, we found that Claude 3.5 Sonnet demonstrated a high level of accuracy in translating Arabic, even recognizing complex prompts in regional dialects and slang,” says Abou Ali. “It also showed exceptional accuracy in identifying profanities and correcting spelling mistakes.” Amazon Bedrock offered more reliable performance than the publicly available generative AI service, while the LLM provided more consistent outputs. “Overall, AWS gave us the technological solution we needed to start delivering Arabic search that was truly built around the individual’s user experience and finding the best place to eat for them in that particular moment using their own Arabic dialect,” states Abou Ali.

As a result of choosing Amazon Bedrock, Places elevated the Arabic search accuracy of its Places UAE app from ordinary to exceptional. It made engagement "more localized," as Abou Ali describes it, giving users the option to search in Arabic using a conversational style of writing that matches their personalities. “Using generative AI on Amazon Bedrock, our app could interpret nuanced phrases in Arabic with an impressive 92 percent accuracy, giving us the confidence to market our Arabic search capabilities,” Abou Ali comments. “Compared to our first attempts with generative AI, our Amazon Bedrock and Claude combination improved search speeds by 50 percent and profanity filtering by the same amount, resulting in a safer and more responsive user experience.”

After the app fully launched in October 2024 with its generative AI capability, the number of Arabic searches has leaped by 1,000 percent as more people downloaded the app following a media campaign.

Daily searches have also risen by nearly 267 percent in the weeks following the full launch, with the total number of dining discovery searches surpassing 100,000. Abou Ali says, “We’re seeing a noticeable increase in user engagement as people see how Places UAE transforms the dining discovery experience regardless of whether your native language is English, Arabic, or one of the local Arabic dialects.”

The Places UAE app is helping people navigate Dubai’s vibrant culinary scene safely, eliminating the need to manually filter their preferences or restrictions. With the app’s growing success, the startup’s cofounders now aspire to expand its current features adopting more AWS services as they target bigger audience. Abou Ali states, “We may be a young startup, but we have ambitious plans for Places UAE. Our experience with Amazon Bedrock demonstrates how AWS provides us with the generative AI tools we need to speed up our progress.” Upcoming features will allow users to upload photos of their dining experiences to the app, with generative AI screening images for inappropriate content before they go live. Abou Ali concludes, “The rapid evolution of AWS generative AI services is ideal for an innovative business like ours that aims to offer a customized, seamless, and secure dining search experience.”