行业动态
dev.to ·
@Samuel Mallo
我构建了一个菲律宾语 AI 诈骗风险分析器：工作原理详解
作者介绍其免费工具 Check Mo Muna，帮助菲律宾用户评估可疑消息。用户可粘贴文本或上传截图，系统先用 OCR 提取文字，再由确定性网络安全规则与 AI 分别评估，结果送入风险引擎生成最终风险分、解释与建议。关键设计是 AI 不做最终决策：风险引擎给确定性规则更高权重，采用 70% 规则 + 30% AI，以提升可预测性同时保留 AI 的上下文理解能力。
AI推荐理由 作者自述构建了面向菲律宾用户的诈骗风险分析工具 Check Mo Muna，采用 OCR + 确定性安全规则 + AI 的多阶段流水线，并明确风险引擎中规则权重 70%、AI 权重 30%，AI 不直接做最终判定。这一权重设计对 Agent/安全类应用有参考价值，但为个人项目自述，无第三方验证或效果数据，需谨慎看待。
62
正文
How I Built Check Mo Muna: An AI Scam-Risk Analyzer for Filipinos
Scam messages are getting harder to recognize, so I built Check Mo Muna, a free Filipino-focused tool that helps users assess suspicious messages before they click or pay.
Users can paste a message or upload a screenshot. The system uses OCR, deterministic cybersecurity rules, and AI to identify potential phishing, fraud, payment scams, social engineering, and other red flags.
The interesting part is that AI doesn't make the final decision.
The analysis pipeline works in several stages:
screenshots are first processed with OCR, then the extracted text is evaluated by deterministic cybersecurity rules and AI. Their results are passed to a risk engine, which produces the final risk score, explanation, and recommendations.
The risk engine gives the deterministic security rules more weight than the AI:
70% Rules + 30% AI
This helps keep the system more predictable while still allowing AI to understand context.
I also spent time testing false positives, especially legitimate security warnings that contain words like OTP, PIN, or password.
The goal isn't to tell users:
“This person is definitely a scammer.”
Instead, Check Mo Muna answers:
“Here are the warning signs. Here's how risky this looks. Here's what you should consider doing next.”
The MVP is live and completely free:
checkmomuna.vercel.app
Built by Samuel Mallo, an AI engineer and cybersecurity professional.
Bago mag-click. Bago magbayad. Check mo muna.