{{ t.msg }}
SwahiliGuard
Fraud Intelligence Platform
LIVE
{{ { connected:'Connected', connecting:'Syncing…', disconnected:'Offline' }[wsState] }}
NLP v2 · Swahili Fraud Detection
Real-time Intelligence Center
Analyzing Tanzanian mobile messages via Telegram in real time
{{ { connected:'WebSocket Live', connecting:'Connecting…', disconnected:'Offline' }[wsState] }}
LinearSVC v2 · 99.4% recall
Total Scanned
{{ fmt(stats.total) }}
messages analyzed
Fraud Flagged
{{ fmt(stats.fraud) }}
threats intercepted
Legitimate
{{ fmt(stats.safe) }}
messages cleared
Live Interception Feed
{{ messages.length }} event{{ messages.length !== 1 ? 's' : '' }}
Awaiting transmissions
Send a message to the Telegram bot and it will appear here instantly over the live WebSocket stream
{{ (msg.username[0] || '?').toUpperCase() }}
@{{ msg.username }}
{{ msg.is_fraud ? 'Fraud' : 'Legitimate' }}
{{ msg.ts }}
{{ msg.text }}
Confidence
{{ pct(msg.confidence_score) }}
Model Status
Operational
ClassifierLinearSVC v2
FeaturesTF-IDF + Char n-grams
Accuracy99.6%
Recall99.4%
Precision99.7%
F1-Score99.6%
ROC-AUC99.99%
DatasetAirtel TZ SMS
Training set10,755 msgs
Threshold0.50
LanguageSwahili (sw-TZ)
BenchmarkBeat RF · 5/5 metrics
SMS Text
→
TF-IDF
→
LinearSVC
→
Signal gate
→
Verdict
Session Telemetry
{{ detectionRate.toFixed(1) }}%
Fraud rate this session {{ stats.fraud }} / {{ stats.total }}
Connection{{ wsState === 'connected' ? 'Connected' : wsState }}
Session uptime{{ uptime }}
Last event{{ lastEvent }}
About

Real-time Swahili fraud detection for a Master's thesis in NLP & cybersecurity.


Messages via the Telegram bot are classified instantly and pushed here over a live WebSocket.

FastAPI
Vue 3
sklearn
WebSocket
Telegram
Docker