Southeast Asian e-commerce giant Shopee has built Compass — a family of specialised, multilingual Large Language Models (LLMs) — with the NVIDIA AI Factory.
Shopee serves hundreds of millions of monthly active users across eight markets in Southeast Asia, Taiwan, and Brazil, facilitating over $136 billion in annual gross merchandise value.
As the platform embedded AI into search, recommendations, fraud detection, and customer service, a structural gap emerged. General-purpose LLMs excelled at English-centric tasks but dropped sharply on the specialized, multilingual work that defines e-commerce in the region.
Shopee’s researchers call this the “jagged frontier”—the same model that scores well on English benchmarks struggles with Indonesian shopping queries, Thai product categorization, and Vietnamese customer intent. To quantify it, the team built EcomEval, the first benchmark dedicated to Southeast Asian ecommerce.
Even leading models showed significant degradation on multilingual tasks, with some open source models dropping below 52% on key categories.
For a platform processing billions of queries per month across six languages, that gap translated directly into lost revenue, slower support resolution, and missed fraud—and the need for a specialized model grew.
According to NVIDIA, Compass outperforms leading general-purpose models for e-commerce tasks, and has already delivered measurable impact.
The tech giant said the LLM achieves 50x faster fraud detection at 90% lower production cost.
Built with the NVIDIA AI Factory, Compass now runs the majority of Shopee’s AI traffic across search, recommendations, fraud detection, parcel recovery and customer service.
In production, Compass now handles the majority of Shopee’s AI traffic. Monthly API tokens grew from 3 billion to 340 billion in eight months—a 113x increase.
NVIDIA said the model powers search and recommendations (driving higher conversion and ad revenue), anti-fraud detection (50x efficiency over manual review, 90% lower processing costs), vision-based parcel recovery in the supply chain (saving millions annually), and LLM-driven customer service that replaced rigid scripted chatbots with context-aware, multilingual interactions.
Shopee’s AI team pretrained Compass-v3, a vertical-domain mixture-of-experts model with 245B total parameters and 71B active per token, designed for Southeast Asian e-commerce.
It was trained on 12T tokens of curated multilingual corpora, including Indonesian, Thai, Filipino, Malay, Tagalog, and Portuguese, as well as large-scale synthetic e-commerce instructions using a mixed-training strategy.
On top of this, Shopee developed a five-stage training pipeline progressing from pretraining through annealing, instruction following, preference alignment, and agentic reinforcement learning.
The pipeline incorporates billion-scale proprietary data mined from product titles, reviews, and ratings, along with novel alignment methods—including OTPO, which dynamically weights tokens by semantic importance, and SPEC-RL, which accelerates on-policy reinforcement learning by 2-3x through speculative rollout reuse.
A mixture-of-experts architecture with Multi-Token Prediction drives further throughput gains.
