AI-Powered Content Detection System banner

AI-Powered Content Detection System

1 devlog
15h 27m 7s

A multimodal AI content detection system that analyses text, images, audio, and video and proactively predicts what misinformation will emerge from official announcements before it spreads.

This project uses AI

Used GitHub Copilot for code completions
Used Claude Opus 4.6 for UI/backend implementation

Demo Repository

Loading README...

lohjo_

Shipped this project!

Before this hackathon started, I knew I wanted to make something deployable, something that any user, even you can use right now. Our problem statement? “How might we design AI-powered solutions that help local and multilingual communities in Singapore assess information credibility, understand context, and make informed decisions — especially during times of uncertainty?”

Our team built a Telegram bot that analyses text, images, audio and video for likely harmful signals. We used locally-trained SEA-LION Guard model for deterministic content detection and ADK-based agent pipelines with Gemini API. However, the journey isn’t over. Future iterations include: group chat invocations, context files using Firecrawl for web-scraping as well as increased security and privacy protections.

I would like to thank the mentors from ClickHouse for validating and enhancing our AI-native workflow and to the organisers at GeeksHacking for volunteering all their time, energy and effort into making such an amazing event!

– Gemini Live Agent Challenge –
Being new to this tech scene, I wanted to jumpstart my portfolio (feels like a taboo saying this) – by extending my project in Gemini Live Agent Challenge. I learnt much of Google Cloud Tech through their YouTube channel. I joined Google’s ADK developer workshops. Watched countless videos on Computer Vision and the way Google’s way-back-home multimodal agent workshop was held changed how I thought about building AI.

Conclusion: This is my starting journey. I want to go on to build projects that I’m passionate about (it might sound like a stretch..) like physics – how elliptic integrals work? compliant mechanisms – how can we compute them? maths – is there a way to generalise quartic factorisation identities for depressed quartic polynomials? However, I think what I’m lacking is a mentor – a technical one. I have embarked on YouthTech’s open-source 6-month commitment programme to hone my skills in open-source contributions, Access Singapore’s commitment to bettering generations of youths. Contact me at: https://linkedin.com/in/lohjohnray or on Slack at @lohjo_

lohjo_

– Result of Two Hackathons –

Our problem statement: “How might we design AI-powered solutions that help local and multilingual communities in Singapore assess information credibility, understand context, and make informed decisions — especially during times of uncertainty?”

Hackomania 2026 (in-person; 24h):

Built a Telegram bot that analyses text, images, audio and video for likely harmful signals. We used locally-trained SEA-LION Guard model for deterministic content detection and ADK-based agent pipelines with Gemini API.
Hackomania Link

Gemini Live Agent Challenge (online):

Orchestrated Gemini Live Agent and Groq to give real-time detection and explanation to fraud/unsafe content. Used Google Cloud servless backend to host website that YOU can use right now!
Devpost Link Here

Future iterations

However, the journey isn’t over. Future iterations include: group chat invocations and increased security and privacy protections and full prediction engine implementation!

Demo Vid Here
^^ go like pleasee

Try it here! Google Cloud Website

Attachment
Attachment
Attachment
Attachment
Attachment
Attachment
Attachment
Attachment
0