Inspiration

Most hackathon demos over-promise intelligence without proving safety, auditability, or failure handling.
ReflexAI was inspired by the need to show how real-time AI systems should be structured first, before adding heavy inference.

What it does

ReflexAI is an event-driven safety intelligence prototype with a deterministic loop, explicit runtime policy control, real voice alerts, and append-only audit logs.
It demonstrates how decisions, policies, and side effects can be separated and verified.

How we built it

The system is built in Python using a LiquidMetal-style orchestration loop.
Decisions are triggered by state changes, passed through a runtime policy layer, and executed via non-blocking side effects such as logging, dashboards, and voice alerts.

A real ElevenLabs API integration generates voice alerts.
A web dashboard renders runtime JSON snapshots.

Challenges we ran into

Running reliable vision inference on CPU-only Windows hardware during a hackathon proved unstable.
To avoid false claims, perception input is explicitly mocked while preserving the system architecture.

Accomplishments that we're proud of

  • A deterministic event loop with audit-grade logging
  • Explicit policy gating with no silent API calls
  • Real, verifiable voice alert generation
  • Honest disclosure of mocked components

What we learned

Clear system boundaries and auditability matter more than feature count.
Honest prototypes build more trust than speculative demos.

What's next for ReflexAI — Event-Driven Safety Intelligence

  • Wiring live vision inference into the decision loop
  • CPU-validated ONNX/OpenCV pipeline
  • Optional cloud persistence

Built With

Share this project:

Updates