Research
* Equal contribution
LionGuard 2
Light-weight, Data-efficient, Multilingual content safety guardrail for Singapore
RabakBench
Benchmarking safety robustness to Singapore's multilingual setting (Singlish, Malay, Mandarin, Tamil)
MinorBench
Benchmarking AI Safety for children in educational settings
Off-Topic Guardrail
Lightweight guardrail to detect off-topic LLM queries
Tinkering
🧧 Gongxi Guru
Practice your Chinese New Year greetings - powered by OpenAI's realtime API
🎙️ Open Notebook LM
Convert any PDF into a podcast episode, using open-source AI models
📰 Daily AI Papers
Summaries auto-generated from HuggingFace's Daily Papers using Gemini and GitHub Actions
🔍 RAGxplorer
Visualise your RAG documents
Selected Writings
Eliciting Toxic Singlish from r1
We discovered that, with just standard prompt-engineering best practices, r1 could generate highly toxic and realistic Singlish content.
From Risk to Resilience: Adding LLM Guardrails From Day 1
7+1 technical tips on how to get started with LLM Guardrails
Building Responsible AI - Why Guardrails Matter
In this post, we discuss why LLM Guardrails are essenital and how we think about designing and implementing them at GovTech
Community
- Cursor Hackathon October 2025
- Amp Meetup August 2025
- Emergency GPT Hackathon August 2025
- Cursor Meetup July 2025
- ICLR 2025 Social - LLMs in the Public Sector April 2025
- smol hackathon for smol models March 2025
- Build Club Global Hackathon June 2024
- AI Engineers Meetup May 2024
- (un)official ai weekend hackathon March 2024
- Pi Day 2024 March 2024
- AI Wednesdays February 2024