SLIDE 1 / 5

Fighting Fake News
Where Students Already Chat

Every day, students share unverified claims and rumours in group chats. We built JarvisClaw as an OSINT verifier that meets them where they already are — right inside Discord and Telegram, requiring zero app installations, zero user training, and zero friction.

How it works — Someone pastes a suspicious claim. The bot pulls out the core facts, checks for red flags, and gives a confidence score. All from a simple !factcheck command in chat.

SLIDE 2 / 5

Smart Routing
Dual-Mode Architecture

Command Mode

Triggered by !analyze !factcheck. Runs claim decomposition, pattern analysis, confidence scoring. Output is clinical, structured, forensic.

💬

Casual Chat Mode

Messages without command prefixes route to the personality layer — witty, adaptive, and conversational. Hard ban on romantic roleplay. Dry sarcasm for shutdowns.

SLIDE 3 / 5

Analysis Pipeline
Triage to Deep Dive

LIGHTWEIGHT TRIAGE DEEP FORENSIC ANALYSIS
Layer 1Source origin, attribution, dates, tone — instant extraction
Layer 2Red flag detection, pattern analysis, confidence breakdown
Layer 3Specific follow-up actions for a human analyst to verify
92%

Accuracy in pulling out core facts

Tested against 50+ real claims from public channels

📈 Live Operations Hub
SLIDE 4 / 5

We Tried to Break It
Here's What Happened

🛡 Jailbreak Defence

Blocked 12+ real jailbreak and prompt injection attempts trying to hijack the bot's memory. Our admin rules stayed locked, and fact-checking ran without interruption.

🔄 Config Corruption

Config file truncated 4 times during testing. Deployed chattr +i — the Linux kernel now blocks any writes to the file at the OS level.

✅ Zero Crashes

Despite aggressive stress-testing and repeated config failures, the gateway and bot completed all fact-check analysis without a single server crash.

SLIDE 5 / 5

Roadmap
Where We're Going

Q1 ✓

Core Gateway & Platform Sync

Built the Node.js core gateway, set up Discord and Telegram synchronisation, and configured basic input filtering. The bot is live and processing real claims.

Q2

Sandboxed Code Execution

Move code execution into fully isolated environments to stop injection risks. No user input will ever touch the host filesystem directly.

Q3

Vector Database Integration

Add a vector database to handle larger text files, search through past claims, and find similar patterns across hundreds of analysed posts.

Q4

Multi-Platform Expansion

Extend beyond Discord and Telegram to a standalone web dashboard for schools and public institutions that need fact-checking without chat apps.

Group 3 OSINT Verifier · June 2026

1 / 5