August 2025 · Southeast Asia (Hybrid)
mAIScam
AngelHack x AI Singapore
Regional AI developer challenge spanning Southeast Asia. 4 Aug – 31 Oct 2025, hybrid format. Organised by AI Singapore with AngelHack as event partner, supported by AWS, Google, NVIDIA, and IMDA. Built using SEA-LION — the region's multilingual open-source LLM. 200+ teams. Placed 2nd overall.
The PAN-SEA AI Developer Challenge 2025 was a regional competition spanning Southeast Asia, organised by AI Singapore and AngelHack, supported by AWS, Google, NVIDIA, and IMDA. 200+ teams. 4 August – 31 October 2025, hybrid format.
We placed 2nd in the Finance Track. Our project — mAIScam — was subsequently featured as an official case study on AI Singapore's SEA-LION website.
Southeast Asia faces a $64 billion cybercrime crisis. Scammers exploit the region's linguistic diversity — launching phishing attacks, fake websites, and social media fraud in 13+ local languages. Current security tools are reactive and culturally blind.
The elderly are disproportionately affected. This wasn't abstract for us — both Shaun and I have family members who fell victim to scams. That's why we built this.
mAIScam is an AI-powered browser extension built on SEA-LION v4 — AI Singapore's multilingual open-source LLM trained on Southeast Asian languages. It provides real-time scam detection across emails, websites, and social media in 13 Southeast Asian languages.
When a user visits a page or opens an email, the extension captures the content and sends it to a backend powered by AWS. SEA-LION v4 analyses the content — text, images, and metadata — and cross-references URLs against PhishTank, validates SSL certificates, and checks for brand impersonation. It returns a risk assessment (High / Medium / Low) with a culturally-appropriate explanation in the user's native language, within seconds.
| Feature | Description |
|---|---|
| Email Protection | Scans Gmail for phishing attempts |
| Website Validation | SSL checks, brand impersonation detection |
| Social Media Analysis | Evaluates Twitter and Facebook posts — text, images, engagement patterns |
| Active Protection | Auto-blocks flagged sites, blurs suspicious content |
| One-click Reporting | Scam reporting to authorities directly from the extension |
| 13 Languages | Culturally-tailored explanations in Southeast Asian languages |
| Layer | Technology |
|---|---|
| LLM | SEA-LION v4 (AI Singapore) |
| Cloud | AWS |
| URL validation | PhishTank API |
| Interface | Chrome browser extension |
| Admin | Real-time threat monitoring dashboard |
| Name | Role |
|---|---|
| Edmund Lin Zhenming | Full-stack engineer — extension frontend, backend integration |
| Shaun Liew Xin Hong | AI & Cloud Engineer — SEA-LION integration, Python, PyTorch, Computer Vision |
Shaun is an AI Research Engineer at SIT × NVIDIA AI Centre specialising in Visual Language Models. This was our second competition together — we'd previously competed at Hackomania 2026 where we won Champion.
Most LLMs are trained predominantly on English data. SEA-LION is specifically trained on Southeast Asian languages and cultural contexts — which means it understands the linguistic nuances that scammers exploit. A phishing email written in Singlish, Bahasa, or Tagalog reads very differently to a native speaker than to a generic English-trained model.
That cultural intelligence is the core of what makes mAIScam work. You can't bolt it on after the fact.
AI Singapore featured mAIScam as an official case study on the SEA-LION website. We're in discussions with AI Singapore for potential research collaboration to enhance SEA-LION's scam detection capabilities, and exploring partnerships with financial institutions, telcos, and government agencies to scale across the region.
The work continues beyond the competition.
We built mAIScam because we've seen what scams do to people we love. The financial loss is real. The shame and confusion that follows is real. Knowing that vulnerable communities across Southeast Asia face this daily — and that the tools to protect them exist but haven't been built yet — was reason enough to spend three months on it.
2nd place, Finance Track. 200+ teams. Built on SEA-LION.