Skip to content
← Back to projects
AI SystemJanuary 15, 2026

USAII Global AI Hackathon — Orca

Human-in-the-loop AI planning workspace using multi-stage agents and context retrieval. Built for the USAII Global AI Hackathon 2026.

PythonAstroReactRedisSupabaseDocker

The Problem

Teams generate ideas, decisions, and action items across chats and documents, but those signals rarely become structured execution plans without heavy manual coordination.

What I Built

Orca is a human-in-the-loop AI planning workspace that ingests team conversations and project artifacts, then produces validated, structured execution plans through multi-stage agents.

How It Works

Multi-stage agents retrieve relevant context, propose plans, and route proposals through human validation gates. Redis handles session state while Supabase stores artifacts and plan history. The Astro + React frontend surfaces agent runs and approval workflows in a single workspace.

Outcome / Results

Finalist at the USAII Global AI Hackathon 2026. Demonstrated that agentic planning with explicit human checkpoints produces more trustworthy outputs than fully autonomous plan generation.