UNESCO Youth Hackathon 2026 Submission|100% On-Device ML

Navigate Social Media Safely with Cultural MIL Literacy

Ramboo is a privacy-first browser extension for real-time hate speech awareness and micro-MIL education. Built specifically for Indonesian, Javanese, and English digital spaces — with zero cloud servers and 100% local processing.

Zero Server Logging
1.2 MB Local ML Engine
Javanese & Indonesian Dialect
Live Extension Preview
RAMBOO ACTIVE
JD
@user_javanese_social• 2h
Identity Attack Detected (0.92)
ID / JV Dialect

Dasar wong ndeso gak nduwe otak, pancen kelompok iku kabeh gak pantes manggon ning kene!

UNESCO MIL Reflection Prompt

Notice the "Us vs. Them" framing stereotyping regional identity. How does language influence social cohesion?

100% On-Device

Zero Cloud Logging

All comment scoring happens inside your browser. No server endpoints, no telemetry leaks.

1.2 MB Model Size

< 5ms Inference Time

Blazing-fast TF-IDF + Multi-Head Logistic Regression compiled into a tight JSON weight bundle.

Multilingual & Dialects

ID, JV & EN Support

Specially tuned for regional Indonesian & Javanese idioms, slurs, and code-mixed online chatter.

UNESCO MIL Framework

Micro-Literacy Nudges

Replaces toxic noise with critical thinking prompts, rhetorical pattern analysis, and reflection.

Interactive On-Device ML Playground

Test Ramboo's Classifier Live

Select a sample comment or type your own to see how Ramboo processes harm detection, sarcasm recognition, and UNESCO MIL reflection prompts in real time.

Max 500 chars
Ramboo ML Classification OutputModel v1.0.0
BLUR (FLAGGED)
Score: 94%
Category: Identity Targeted Derogation

Dasar wong ndeso gak nduwe otak, kelompok iku pancen gak pantes manggon ning kene!

UNESCO Micro-MIL Reflection

Notice the use of regional identity slurs to create societal divisions. How does stereotyping affect digital community safety?

Pattern Identified: Us vs. Them Framing & Dehumanization
Designed for Everyday Internet Users

Why Ramboo Sets the New Standard

Combines local machine learning, cultural empathy, and Media & Information Literacy (MIL) into a non-censoring, user-controlled browser extension.

Performance & Privacy

100% On-Device ML Engine

Executes TF-IDF vectorization and multi-head logistic regression locally in under 5 milliseconds. No text ever leaves your browser window.

User-Controlled & Privacy First
Linguistic Nuance

Indonesian & Javanese Dialect Awareness

Specifically trained on regional expressions, code-mixed dialects (Surabaya, Yogyakarta, Semarang), and localized slang that standard classifiers miss.

User-Controlled & Privacy First
UNESCO Framework

Micro-MIL Educational Nudges

Instead of blunt censorship, Ramboo presents contextual reflection cards that build long-term critical thinking and media literacy skills.

User-Controlled & Privacy First
User Empowerment

Custom Sensitivity Sliders

You control what gets flagged. Adjust sensitivity per harm category (identity attacks, harassment, profanity, political debate) from 0 to 100.

User-Controlled & Privacy First
Multi-Platform

Granular Domain Control

Seamless integration with X/Twitter, YouTube, Reddit, Instagram, Facebook, and major Indonesian news portals (Kompas, Detik, Tempo).

User-Controlled & Privacy First
Smart Disambiguation

Sarcasm & Counter-Speech Aware

Distinguishes between genuine identity attacks, sarcastic mock praise, and constructive counter-speech to prevent annoying false positives.

User-Controlled & Privacy First
Open Source & Transparent Architecture

How Ramboo Operates 100% On-Device

By compiling our machine learning classifier into a compact JSON weight matrix, Ramboo performs inference directly in your browser without external servers.

Inference Execution Pipeline

01Chrome / Firefox MV3

DOM Adapter Selection

Content script observes comment nodes on X, YouTube, Reddit, Instagram, Facebook, and news portals. Skips strings < 5 characters.

`src/content.ts`
021.2 MB Local Weights

On-Device TF-IDF Engine

Extracted text is tokenized into word n-grams and evaluated against multi-head logistic regression weights in < 5 milliseconds.

`model_weights.json`
03Isolated Shadow DOM

Context Card & Micro-MIL

Applies CSS blur, surfaces category scores, and injects UNESCO reflection prompts via isolated Shadow DOM elements.

`src/components/ContextCard.tsx`

ramboo-extension

TypeScript + React

The Manifest V3 browser extension installed by users. Contains platform adapters, Shadow DOM context cards, onboarding calibration, and settings dashboard.

License: MITView Repository

ramboo-ml

Python + scikit-learn

The machine learning training pipeline. Generates synthetic corpora, trains multi-head classifiers, evaluates counter-speech accuracy, and exports JSON model weights.

License: MITView Repository
First-Install Onboarding Calibration

Tailored to Your Personal Sensitivity Boundaries

When you first install Ramboo, a 10-scenario calibration quiz builds your personal Content Preference Profile. You can recalibrate or adjust sliders anytime.

1Calibration Scenario 1 of 10
Exclusionary Language
"Orang-orang dari pulau sebelah emang selalu bikin onar, mending dipulangin aja semua!"
Frequently Asked Questions

Everything You Need to Know

No. Ramboo runs 100% on-device inside your browser window. All ML scoring, text tokenization, and context overlays execute locally using a compact 1.2 MB bundled model matrix. There are no remote API endpoints or tracking servers.