Lim Cheng Yee

AI Product Leader

8+ years building and governing AI systems at scale — from fine-tuning Singapore's first sector-specific LLM to architecting responsible AI governance infrastructure for 1B+ users. I bridge deep technical execution with cross-functional programme leadership across engineering, policy, and senior stakeholders.

Responsible AI & Governance LLM Fine-tuning & RLHF Privacy-Preserving AI Data Strategy Solution Architecture
LinkedIn Get in touch

Featured Work

Key Projects

From model development to responsible AI adoption — across government, enterprise platforms, and national-scale deployments.

01 IMDA · SAL · AWS · Sept 2024

GPT-Legal

Singapore's First Sector-Specific LLM

Led end-to-end development of a domain-adapted LLM for Singapore's legal sector — from model training and responsible AI tooling to deployment on LawNet 4.0 serving 75% of the country's practising lawyers.

  • →Domain-specific SFT on Amazon SageMaker; RLHF pipeline with Singapore Courts legal clerks; DPO for alignment to legal formatting standards
  • →Productionised hallucination detection with a 90% accuracy threshold enforced before every user-facing release
  • →End-to-end data lineage (S3/SageMaker) for compliance across public and proprietary legal data during training
  • →Aligned SAL, AWS, and Singapore Courts across a multi-quarter roadmap; presented to the Chief Justice of Singapore
15,000+
judgments summarised
hrs → mins
research time
~10,000
lawyers served
LLM Fine-tuning RLHF / DPO Responsible AI Hallucination Detection AWS SageMaker Data Lineage
GovInsider Article AWS Case Study
02 TikTok | ByteDance · 2024–Present

GenAI Safety-as-a-Solution

Horizontal AI Governance Platform · 1B+ Users

Architected a unified AI governance platform to replace six siloed, product-specific safety stacks across chatbots, image/video generation, code, ads, and ModelArk with a single authoritative standard.

  • →Defined unified GenAI risk taxonomy and assessment frameworks replacing ad-hoc, product-siloed safety approaches
  • →Architected reusable moderation pipelines across text, image/video, code, and model-level safety surfaces
  • →Built and led the GenAI Safety Task Force — aligning Product, Policy, Algo, DS, and Ops across timezones for 24/7 coverage
  • →Standard adopted cross-platform; consistent enforcement across 6+ verticals for 1B+ users globally
1B+
users covered
6+
product verticals
1 standard
cross-platform
AI Governance Responsible AI Risk Frameworks Platform Architecture Programme Management XFN Leadership
03 TikTok | ByteDance · 2024–Present

Agentic Moderation & Model Evaluation

LLM/VLM Agentic Pipelines · Evaluation at Scale

Replaced fragmented keyword-filter pipelines and manual labelling workflows with agentic AI systems and a productionised evaluation lifecycle — driving measurable gains in speed, cost, and model quality.

  • →Redesigned cross-product moderation to a unified LLM/VLM agentic workflow with prompt engineering and image tagging models
  • →Productionised AI-powered autolabelling pipeline achieving 80% precision/recall vs. human-curated safety standards
  • →Built end-to-end safety evaluation lifecycle with leadership leaderboard; advised on model quality vs. GPU cost trade-offs
↓60%+
moderation load
3×
labelling cost reduction
↑75%
faster iteration
Agentic AI LLM / VLM Model Evaluation MLOps Autolabelling GPU Optimisation
04 Prime Minister's Office · 2019–2022

Centre for Citizen Insights

Privacy-Preserving Analytics · National Scale

Built Singapore's Cabinet-level policy analytics platform with privacy-enhancing techniques at its core — enabling cross-agency data collaboration without exposing citizen PII.

  • →Implemented hashing, fuzzy matching, and aggregated learning to protect citizen PII across cross-agency data joins — enabling analysis without direct data sharing
  • →Explored federated learning and synthetic data generation as privacy-enhancing approaches for future cross-agency collaboration
  • →ML clustering and behavioural analytics engine; Tableau data product for cross-agency teams; MVP delivered in 4 months
  • →Insights presented directly to the Singapore Cabinet; informed 2 national policy pivots with >85% public alignment
↓500%
decision latency
↓75%
ad-hoc requests
Cabinet
presented directly
Privacy-Enhancing Tech Hashing & Fuzzy Matching Federated Learning Synthetic Data Data Strategy ML Analytics

Capabilities

Expertise

Technical depth and strategic programme execution — not as trade-offs, but as a single capability.

AI Solution Architecture

LLM Fine-tuning & Post-training
SFT, RLHF, DPO, domain adaptation, alignment
Model Evaluation & MLOps
Evaluation pipelines, autolabelling, quality/cost trade-offs, GPU optimisation
Agentic AI Systems
LLM/VLM agentic workflows, prompt engineering, multi-modal pipelines
Privacy-Enhancing Technologies
Federated learning, synthetic data generation, hashing, fuzzy matching
Cloud Architecture
AWS (SageMaker, S3, EC2), data lineage, compliance-ready ML pipelines

Product Roadmapping

0→1 Product Development
Concept to production; sector-specific AI deployment; platform strategy
Agile Delivery
Certified CSM & CSPO; sprint planning, MVP scoping, iterative delivery
User Research & Discovery
Qualitative synthesis, user archetypes, insight-to-strategy translation
GenAI Platform Strategy
Horizontal platform design, safety-by-design, GTM planning
Data Product Development
Cross-agency analytics platforms, self-serve data products, Tableau

Programme Management

Cross-Functional Leadership
Engineering, Policy, Legal, Ops alignment; task force management across timezones
Multi-Stakeholder Delivery
Multi-quarter roadmaps; government-industry partnership management
Executive Stakeholder Management
Cabinet, C-suite, and senior government leadership briefings and alignment
Vendor & Partnership Management
AWS, Singapore Academy of Law, tech industry and public sector partnerships
Regulatory & Policy Engagement
Digital Economy Agreements, ASEAN frameworks, compliance-driven delivery

Responsible AI

AI Governance Frameworks
Risk taxonomy design, safety standards, cross-platform policy adoption
Hallucination Detection & Output Validation
Safety-by-design, accuracy thresholds, discrepancy flagging before deployment
Privacy & Data Ethics
PII handling in ML workflows, privacy-by-design, data lineage for compliance
AI Risk Assessment
Risk categorisation, pre/post-launch safety evaluation, trend monitoring
Safety Engineering
Output filtering, prompt sanitisation, content safety pipeline design

Career

Experience

Across industry AI, government technology, and national policy.

AI Product Manager, Trust & Safety

TikTok | ByteDance

2024 – Present · Singapore

Spearheaded GenAI Safety-as-a-Solution — a horizontal content safety platform for a 1B+ user platform. Led agentic moderation systems, AI-powered autolabelling pipelines, and model evaluation lifecycle across 6+ GenAI product verticals including chatbots, image/video generation, code, ads, and ModelArk.

Senior Product Manager, AI Engineering

Infocomm Media Development Authority

2022 – 2023 · Singapore

Led GPT-Legal — Singapore's first sector-specific LLM — from domain-specific fine-tuning and RLHF pipeline to responsible AI deployment on LawNet 4.0 in partnership with the Singapore Academy of Law and AWS. Summarised 15,000+ court judgments; serves 75% of Singapore's practising lawyers.

Senior Manager, Policy Analytics

Prime Minister's Office Strategy Group

2019 – 2022 · Singapore

Established the Centre for Citizen Insights under the Head of Civil Service. Built ML-powered analytics platform informing Cabinet-level policy decisions; delivered MVP in 4 months and grew user pipeline 50%+ annually. Insights informed 2 national policy pivots with >85% public alignment.

Assistant Manager, Industry Division

Ministry of Communications and Information

2017 – 2019 · Singapore

Expanded Singapore's TechSkills Accelerator by $145M (13,000 new training places). Led user research, synthesised insights into policy archetypes for Singapore's global tech talent strategy, and reviewed Digital Economy Agreements and ASEAN digital connectivity frameworks.

Background

Education & Credentials

2016 – 2017 · United States

MA, Social Sciences

University of Chicago

Certificate in Computational Science & AI

2013 – 2016 · United Kingdom

BSc Economics

University of Warwick

Public Service Commission Scholar — one of 70 recipients

Certifications & Recognition
Certified Scrum Product Owner
CSPO · Scrum Alliance · 2022
Certified Scrum Master
CSM · Scrum Alliance · 2021
Public Service Leadership Programme
Selected top talent · Singapore Civil Service · 2017–2023

Let's connect

Based in Singapore.

limchengyee@outlook.com LinkedIn