Decide

• 

  Decide

About DECIDE

Inspiration.
Real-world operations don’t fail in isolation — they fail amid parallel conversations, partial signals, and time pressure. I wanted to explore how AI agents could coordinate context, not just answer questions, while keeping humans firmly in control.

What I built.
DECIDE is a real-time incident response system where AI agents aggregate signals (system state, time, location, cross-team updates), continuously reassess severity, and recommend an escalation action — which a human operator authorizes.

How it works.
Multiple simulated inputs feed an agentic orchestration layer (deterministic rules + LLM reasoning). Claude is used to summarize live coordination and surface decision-relevant signals, while the UI reflects risk visually (severity band + map).

What I learned.
The hardest part isn’t reasoning — it’s context orchestration: deciding what information matters, when to re-evaluate, and how to present decisions so humans can act confidently in seconds.

Challenges.
Balancing automation with human authority, keeping the interface glanceable under stress, and making agent behavior feel alive rather than static.

Built With

• claude
• openstreetmap