Generative AI in Production: From Proof of Concept to Sub-100ms Inference
A preview of Dr. Cheryl Wong’s keynote exploring low-latency LLM orchestration and multilingual Southeast Asian foundational models.
While experimenting with LLM APIs in Jupyter notebooks is straightforward, serving generative AI models to millions of concurrent production users with sub-100ms latencies and strict data governance is an entirely different engineering discipline.
On Day 2 of DevSummit Singapore 2026, Dr. Cheryl Wong (Principal AI Researcher at AI Singapore Labs) will dissect real-world production architectures deployed across ASEAN enterprise environments.
Architectural Topics Covered
- vLLM & Speculative Decoding: Squeezing maximum token throughput out of constrained GPU server clusters.
- Multilingual Embeddings: Overcoming semantic drift across Malay, Bahasa Indonesia, Tagalog, and Singlish vernacular.
- Hybrid Cache Topologies: Semantic caching with vector databases to eliminate redundant GPU inference costs by up to 65%.
Join us on Day 2 in the Suntec Level 3 Keynote Hall to explore the code and infrastructure powering the next generation of AI-native applications.
Join us at DevSummit Singapore 2026
Get your pass today and connect directly with keynote speakers and peers.