AI | Machine Learning Architect

Feb 2026 - Present

Spearheaded the architecture and deployment of autonomous AI agents using LangGraph and multi-agent 'deep agent' patterns, automating complex financial support workflows and achieving a 25% reduction in customer support ticket volume.
Designed high-fidelity agent architectures for advanced reasoning and tool-use, enabling autonomous navigation of internal APIs and documentation to resolve high-intent support cases with minimal human intervention.
Engineered a unified intelligence layer for the Airwallex AI Assistant, integrating disparate data-fetching and analytics capabilities to automate 20% of common data queries and significantly reduce operational latency.
Defining the department’s Generative AI vision, establishing standardized evaluation frameworks and guardrails for 'deep agent' patterns to ensure safe, deterministic outcomes in a highly regulated fintech environment.
Acting as the technical bridge between product and engineering to identify high-impact LLM opportunities, while mentoring senior engineers on the productionization of agentic systems.

Aug 2025 - Jan 2026

Engineered an end-to-end video summarization and semantic search pipeline for a government-scale surveillance project, integrating Vision LLMs (VLMs) and Computer Vision models to transform raw footage into searchable, actionable metadata.
Designed and implemented a hybrid RAG and GraphRAG framework to enable complex entity-relationship queries across unstructured surveillance data, allowing for deep-link analysis of detected subjects and events.
Optimized large-scale VLM deployment using vLLM and DeepStream plugins, significantly increasing inference throughput and enabling real-time processing of concurrent high-definition video feeds.
Developed a framework that bridged traditional object tracking with generative reasoning, automating natural language incident reporting and situational awareness for high-security environments.
Scaled the organization's AI capability by mentoring and training traditional 10+ software engineers in ML/AI proficiencies, including but not limited to object detection/tracking, prompt engineering, model evaluation, and vector database management.
Engineered the CI/CD pipeline for the AI platform, automating model deployment, versioning, and inference testing to ensure high reliability in a mission-critical government environment.

Apr 2024 - Aug 2025

Lead AI-driven content risk control for TikTok LocalService, implementing cutting-edge LLM and GraphSage models.
Developed spam perception algorithms (Leiden clustering, content classification) to detect fraudulent content.
Built real-time behavioral sequence models, reducing fraud/spam activities by 90%.
LoRA fine-tuned multiple LLM models (Qwen2.5/Gemma3/Llama3.2) for content violation prediction. Reduced violative POI reviews by 60% at launch.
Long term technical strategy planning & execution for Tiktok LocalService overall risk control.
Collaborated with cross-functional teams (XFN) to design anti-fraud policies and risk mitigation strategies.
Mentored junior/senior data scientists, fostering AI innovation and knowledge sharing.

Nov 2021 - Apr 2023

Developed and deployed high-precision fraud detection ML models, reducing risk in the TikTok Live ecosystem by 70%+.
Designed real-time fraud detection pipelines to monitor and flag suspicious activity with >90% accuracy.
Automated behavioral sequence modeling using deep learning (LSTM, Transformers) to detect scam patterns.
Partnered with Trust & Safety teams to prevent fraudulent activities across TikTok’s global ecosystem.
Trained and mentored new hires in advanced ML techniques and AI deployment.

Nov 2019 - Jun 2021

Led APAC network planning initiatives, optimizing mobile tower placement and/or fiber routing using state of the art AI/ML solution.
Built automated data pipelines to generate network performance reports for global telecom partners.
Developed ML-based predictive models for infrastructure expansion and network optimization.

Oct 2018 - Oct 2019

Productionized fully automated ML models for repurchase propensity model and lead recommendation model.
Applied NLP and text mining to analyze call center data, extracting key insights for business decisions.
Automated geocoding and location-based analytics to improve agent allocation efficiency.
Mentor and assist junior data scientist in various innovation projects.

Sep 2016 - Sep 2018

Built churn and NPS prediction models to maximize ROI on network investments.
Developed network congestion forecasting models, improving customer experience by 20%.

Apr 2014 - Sep 2016

Used geospatial analytics to reduce customer complaints by 25% through network optimization.
Deployed mobile coverage solutions for major events like Formula 1 and Moto GP.

Goh Yee Chern