LiveOps - Generative UI Global Hackathon: Agentic Interfaces
AI Tinkerers - Durango, MX
Hackathon Showcase Finalist

LiveOps

Team led by Alfredo Flores Garcia, an ITJ Software Engineer specializing in Keycloak SPIs, OAuth 2.1, and high-concurrency Java/Kotlin authentication infrastructure modernization.

1 member

LiveOps is a generative NOC (Network Operations Center) dashboard that replaces static incident runbooks with AI-driven, dynamically-generated interfaces. When a service goes down, the interface doesn’t just display alerts — an AI agent analyzes the incident context and generates the appropriate UI in real time.
How it works: A health poller monitors Keycloak on Railway via SSE. When an incident is detected, a CopilotTask fires automatically — no chat input, no prompt. Gemini 2.0 Flash receives the incident payload (error rate, impacted users, last healthy timestamp) and calls the show-remediation A2UI action, which renders a RemediationCard containing a situation analysis and context-specific action buttons. A database deadlock surfaces different buttons than an OOM error — the interface adapts to the problem, not the other way around.
Why this goes beyond chat: Traditional AI copilots live in a sidebar — you type, they respond with text. LiveOps uses CopilotKit’s A2UI (Agent-to-UI) pattern via useCopilotAction to let the agent render live React components directly into the dashboard. The AI doesn’t describe what to do; it builds the interface for doing it. Users click AI-generated action buttons that trigger real Railway API deployments, with an approval flow for safety.
Stack: CopilotKit 1.57.1 (@copilotkit/react-core for A2UI actions, CopilotTask for programmatic agent triggers, @copilotkit/runtime + GoogleGenerativeAIAdapter on the backend), Gemini 2.0 Flash as the reasoning model, React 18 frontend, Express backend with SSE for real-time events, Railway GraphQL API for infrastructure operations, and TypeScript throughout.
Originality: The core insight is that generative UI for ops isn’t a chatbot — it’s a dashboard that reconfigures itself per-incident. The same CopilotTask architecture that today proposes actions for human approval scales to autonomous remediation: the agent generates the interface, the human approves, and eventually the agent executes directly. The generative UI becomes the audit trail for self-healing infrastructure.

Only worked on the demo app to be tested: https://github.com/codedrifter-mx/myAwesomeApp, and the railway account and infra setup

A2UI AG-UI AI Tinkerers CopilotKit GLM 5.1 Google DeepMind Opencode Opendesign Railway Universidad Tecmilenio deepseek 4 flash

LiveOps main repo

Summarizing URL...

myAwesomeApp to be used for LiveOps actual product

Summarizing URL...

LiveOps live demo

Summarizing URL...