/ CASE STUDY
ORCA
JAN 2026 — PRESENTPROJECT LEAD — ARCHITECTURE, BUILD, DELIVERY
ORCA is a hub orchestrator for federated AI agents: one chat interface, where a lead agent plans each request, delegates to specialised member agents, and streams the synthesised answer back in real time. It runs inside a private network, so what you're seeing below is an illustrative miniature — fictional agents, fictional data, recreated at small scale to show the experience. It is not the production system.
/ LIVE DEMO
ILLUSTRATIVE DEMO · FICTIONAL DATA
orca.example — demo
/ MULTI-AGENT CHAT
One chat. A team of agents.
Ask once — a lead agent plans, delegates to specialists, and answers with the receipts.
/ UNDER THE HOOD
One chat, many specialists
A lead agent plans the request, picks the right member agents for each part, and synthesises their results into a single answer — the user never has to know which agent does what.
Transparent orchestration
Every delegation, tool call, and reasoning step streams to the browser as it happens, so the work is inspectable rather than a black box.
Streaming-first architecture
A typed event protocol over server-sent events drives the whole UI — text, tool activity, progress, and interactive outputs all arrive as a live stream.
Led end to end
I lead the project across the React frontend, the Python agent backend, and the infrastructure between them — from architecture to delivery.
/ FURTHER READING