I

IRIS

Information Risk Intelligence System

◐ BETA

Information risk intelligence.

Automated monitoring and classification of hate speech, disinformation and violent-extremism narratives across social media and online platforms. Analysts get a map-based timeline and a chat assistant for triage.

Request a demo
IRIS screenshot 1: East Africa disorder and hate-speech monitor: map and event timeline
East Africa disorder and hate-speech monitor: map and event timeline
IRIS screenshot 2: Analyst chat and narrative-family triage panel
Analyst chat and narrative-family triage panel

Capabilities

What it does.

Capabilities listed here are present in the current build. Performance figures, model cards and annotation protocols available under NDA.

  • ·Four fine-tuned classifiers across East African hate-speech and polarisation use cases
  • ·Multilingual: Arabic, Somali, English, Swahili, Amharic, Oromo, Tigrinya, Kinyarwanda and growing
  • ·Runs on standard CPU infrastructure. No GPU needed for day-to-day inference
  • ·PII redaction on by default
  • ·Narrative-family timelines with P1 (critical) to P4 (low) triage
  • ·Built-in analyst chat for narrative search and briefing support
Built on
Python·HuggingFace Transformers·XLM-RoBERTa·Next.js·Supabase·D3.js

Classifier performance per language, model cards and annotation protocol available under NDA.

← All Optics toolsTalk to the studio →