Skip to content

CrisisCore

CrisisCore is a multimodal, multi-agent disaster response system built on Claude Opus 4.6.

Code Video Cerebral Valley

Core trick

Six agents score each incoming report for confidence, and a debate step keeps contradicting sources visible instead of overwriting one with another.

Tags

agentsvision

Description

CrisisCore is a multimodal, multi-agent disaster response system built on Claude Opus 4.6. Six specialized agents process chaotic emergency signals — drone imagery, 911 calls, radio dispatches, social media, into a live situation graph with confidence-scored incidents, detected contradictions, and resource allocation recommendations with explicit tradeoffs. The key differentiator is epistemic transparency: when sources contradict each other, CrisisCore surfaces the disagreement rather than hiding it. A structured Debate Room lets four Claude agents argue live (Defender, Challenger, Rebuttal, Synthesis), producing confidence-scored verdicts for human operators. The system includes three interfaces: a real-time command dashboard with mapping and evidence flow visualization, the multi-agent debate room for resolving contradictions, and a conversational co-pilot with full situational awareness. Every decision is logged with reasoning, confidence levels, and operator attribution.
ArjunBalaji79/Claude-Opus-4.6-Hackathon★ 00 forksPythonMITCreated Feb 16, 2026 · last push Jul 7, 2026

README excerpt

# CrisisCore

**Multimodal, Multi-Agent Disaster Response Coordination with Epistemic Transparency**

> When a 6.8 earthquake hits, contradictory reports flood in simultaneously. A satellite shows a bridge intact. A first-responder says it collapsed. Social media is hysterical. CrisisCore uses 6 specialized Claude agents to process the chaos, surface the contradictions, and let human operators make auditable decisions with explicit tradeoffs.

Built with **Claude Opus 4.6** for the Anthropic Hackathon. Selected as one of 500 builders across the world for the final round. Demo: https://youtu.be/v__P93xmKig

---

## The Core Insight

Most disaster AI hides uncertainty. CrisisCore **surfaces it**.

When agents disagree, the system doesn't pick a winner — it shows the disagreement, lets agents debate it live, and gives the human operator everything they need to decide.

---

## How It Works

```mermaid
flowchart LR
    subgraph Signals["Incoming Signals"]
        IMG["🛰 Drone/Satellite\nImages"]
        AUD["📻 911 Calls\nRadio Dispatches"]
        TXT["📱 Social Media\nOfficial Reports"]
    end

    subgraph Agents["Claude Agents"]
        VA["VisionAgent\n(damage assessment)"]
        AA["AudioAgent\n(transcript analysis)"]
        TA["TextAgent\n(credibility scoring)"]
    end

    IMG --> VA
    AUD --> AA
    TXT --> TA

    subgraph Graph["Situation Graph"]
        INC["Incidents\n+ confidence scores"]
        CON["Contradictions\ndetected across sources"]
        ACT["Actio

Team