Boosts items listed per seller by 20% with an intuitive, AI-driven experience powered by Gemini
Reduces median listing time to less than two minutes by automatically generating listings with Gemini
Cuts on-call workload for key compute clusters to zero by migrating to more performant C3D Compute Engine instances
Automates 90% of translation needs and eliminates subscription costs with a custom workflow built on the Gemini API
Carousell, a multi-category classifieds and recommerce marketplace, simplifies selling and buying. Leveraging Google Cloud solutions like Gemini and Vertex AI, Carousell enhances user experience and automates internal processes from translation to HR.
Carousell, a multi-category classifieds and recommerce marketplace, simplifies selling and buying. Leveraging Google Cloud solutions like Gemini and Vertex AI, Carousell enhances user experience and automates internal processes from translation to HR.
A mission of simplicity meets the complexities of scale
Carousell was founded on a simple premise: to make buying and selling intuitive and accessible for everyone. The gateway to this experience is the "Sell Form," which enables users to list their items. As a multi-category platform, providing detailed and relevant attributes for products is crucial for helping them get discovered and sold quickly. However, as Carousell added more of these features over its 13-year history, the form became increasingly complex, creating friction in the user experience.
Carousell knows that a seamless experience is also founded on creating a safe and trusted marketplace for all. To protect its community, Carousell proactively detects and prevents fraudulent activity from user chats, such as attempting to take transactions off-platform with contact details or malicious links. The company’s custom-built AI model for analyzing this behavior, however, required a lengthy development cycle and was difficult to scale cost-effectively due to the platform's high volume of conversations.
On the core technology front, the platform’s N2D machines struggled to keep up with Carousell's growing workloads, causing instability that consumed nearly 30% of the on-call infrastructure team's time, pulling them away from product innovation. Internally, critical processes like content translation for multiple markets and handling routine HR queries remained manual and costly, creating a significant operational drag on the organization.
From our early days, our focus was on having a stable, scalable cloud platform. As we've grown, our challenges have become more diverse. We needed a unified ecosystem that could solve for AI-driven user experiences, data intelligence, and core infrastructure simultaneously. Google Cloud provides that for us.
Rajath Ramesh
Group Director, Product and Platform Engineering, Carousell
Rather than using different vendors for each challenge, the company chose to build its solutions on Google Cloud's integrated and scalable platform. "From our early days, our focus was on having a stable, scalable cloud platform,” says Rajath Ramesh, group director of product and platform engineering at Carousell. “As we've grown, our challenges have become more diverse. We needed a unified ecosystem that could solve for AI-driven user experiences, data intelligence, and core infrastructure simultaneously. Google Cloud provides that for us."
Reimagining the selling experience with multimodal AI
To solve the complexity of the "Sell Form," Carousell used Gemini to create an entirely new "List with AI" feature. The solution allows a seller to initiate a listing with either a short video or a series of photos. With on-screen prompts for guidance, the user can simply capture their item, and Gemini’s multimodal capabilities process the video, audio, and imagery in real time.
We were excited by Gemini because it was the only quality, multimodal model that could process video, audio, image, and text together. This was key, as it allowed us to avoid the complexity and user-perceived latency of stitching multiple solutions together.
Rajath Ramesh
Group Director, Product and Platform Engineering, Carousell
"We were excited by Gemini because it was the only quality, multimodal model that could process video, audio, image, and text together,” Ramesh says. “This was key, as it allowed us to avoid the complexity and user-perceived latency of stitching multiple solutions together."
The AI then automatically populates the entire listing, from the title, description, and category to the price and condition attributes. It even offers several description styles that the user can select with a single tap — from simple and direct to more sales-driven. This new workflow shifts the user’s role from actively drafting content to simply reviewing it, making the listing process significantly easier.
With this new intuitive, AI-driven experience, the median listing time has been reduced to less than two minutes, contributing to a 20% increase in items listed.
With the initial launch of this new AI-driven experience, the median listing time has already been reduced to less than two minutes, contributing to a 20% increase in items listed. Carousell continues to make improvements in an effort to reduce this time even further.
Building an agile, resilient platform to power growth at scale
Carousell’s previous custom-built model for analyzing user chats for fraud covered only 10-20% of conversations and was difficult to scale cost-effectively. To analyze its full dataset, the team now leverages the seamless integration between BigQuery, where its chat data resides, and Vertex AI to run bulk inferences with Gemini models.
"The seamless integration between BigQuery and Vertex AI allowed us to move much faster," Ramesh says. "We can now apply AI for bulk processing across our entire dataset without having to build a complex new data pipeline from scratch."
Previously, improving the model required a three-week process of data collection, retraining, and deployment. Now, the team can rapidly refine and iterate on its fraud detection methodology in just a couple of days.
On the infrastructure front, as Carousell's platform traffic grew, its compute-heavy workloads began to outpace the capabilities of its older N2D machines. This led to random performance spikes and platform instability, which consumed nearly 30% of the on-call engineering team's time.
To solve for both performance and reliability, Carousell migrated these workloads to C3D machines, a newer, more performant class better suited for high-volume use cases. The migration immediately eliminated the source of instability, cutting the related on-call burden for engineers to zero and reducing the cost of the compute cluster by 10%.
The seamless integration between BigQuery and Vertex AI allowed us to move much faster. We can now apply AI for bulk processing across our entire dataset without having to build a complex new data pipeline from scratch.
Rajath Ramesh
Group Director, Product and Platform Engineering, Carousell
Driving company-wide efficiency with automation
To serve a diverse, multi-language region like Southeast Asia, Carousell previously relied on paid, external contracts for translating content — a costly and time-consuming process. To improve efficiency, the company built a custom, automated translation workflow using the Gemini API. This new system now handles 90% of Carousell’s translation needs automatically, with an operations team reviewing only the edge cases where confidence is low, eliminating the need for paid subscription services entirely.
Operating across seven markets also presented a challenge for Carousell’s HR team, which spent considerable time addressing repetitive queries about leave, holidays, and claims procedures. Using Gemini, Carousell developed an internal AI-powered Slack bot that securely accesses company policies stored in Google Drive to provide employees with instant, self-service answers.
This has greatly reduced the team's routine workload, freeing them to concentrate on higher-impact initiatives. This holistic approach is part of a broader company strategy. "We apply AI to our own organization," Ramesh says. "By automating internal processes like translations and HR queries with Gemini, we can operate more efficiently and empower our teams to do their best work."
We apply AI to our own organization. By automating internal processes like translations and HR queries with Gemini, we can operate more efficiently and empower our teams to do their best work.
Rajath Ramesh
Group Director, Product and Platform Engineering, Carousell
Building tomorrow's frictionless online selling
Carousell’s journey with AI is accelerating. The company has a bold ambition to make its "List with AI" feature even faster, with a goal to get listing times to under 45 seconds for all users. To achieve this, Carousell plans to explore the latest Gemini models to further enhance the quality of its listings. This includes using AI to cleanse user-generated data and creatively enrich it with stylistic attributes — such as "minimalist" or "Scandinavian" — to improve discovery for buyers.
In addition to speed and quality, personalization is high on the team's agenda. “We're working to make listings faster and more personal,” Ramesh says. “We're exploring how AI can reflect our sellers' unique character, moving beyond robotic descriptions to create a more human and engaging marketplace."
From an engineering perspective, the company is actively exploring agentic coding tools like Jules to automate complex internal tasks, such as analyzing system architecture reports and automatically generating pull requests for code upgrades. The goal is to transform multi-day engineering efforts into a few hours of simple validation, freeing up developers to focus on building the next generation of features.
Carousell Group is the leading multi-category platform for secondhand goods in Greater Southeast Asia, with a mission to make secondhand the first choice. Founded in 2012, the group serves tens of millions of monthly active users across seven markets. Believing in the power of possibilities, Carousell aims to inspire everyone to start buying and selling to make more possible for one another.
Industry: Retail
Location: Singapore
Products: Gemini, Vertex AI, BigQuery, Cloud Storage, Compute Engine, Google Kubernetes Engine, Google Workspace
