Avirat AI Platform

• Provide a dashboard for configuring and monitoring AI-driven workflow automation
• Support prompt-driven workflows with dynamic, unpredictable response shapes
• Process webhook-triggered events and reflect them in the UI in real time
• Support enterprise authentication and role-based administration
• Surface usage analytics for automated workflows

The dashboard treats every AI-driven feature as a producer of events into one real-time update stream, rather than each module maintaining its own separate polling or refresh logic. Prompt-driven workflows and webhook-triggered actions both feed into that same update mechanism, so the dashboard reflects new activity consistently regardless of which module triggered it.

AI Integrations, Prompt Orchestration, and Webhook Processing were each built as isolated feature modules with a consistent internal shape (API layer, Redux slice, components), so adding a new AI-driven capability meant following an established pattern rather than inventing a new one each time. A shared API layer normalized how every module talked to the backend, which mattered more here than on more typical CRUD apps because AI responses are inherently less predictable in shape and timing.

src/
├─ app/
│  ├─ dashboard/page.tsx
│  ├─ workflows/[id]/page.tsx
│  └─ analytics/page.tsx
├─ features/
│  ├─ ai-integrations/
│  │  ├─ aiSlice.ts
│  │  └─ components/
│  ├─ prompt-orchestration/
│  ├─ webhooks/
│  │  ├─ webhookSlice.ts
│  │  └─ use-realtime-events.ts
│  ├─ analytics/
│  └─ administration/
├─ components/ui/
├─ lib/
│  ├─ api-client.ts
│  └─ auth/

Each AI-driven module owns its own Redux Toolkit slice for its specific workflow state, while incoming real-time events (from webhooks or long-running prompt executions) are dispatched through a shared events slice that any module can subscribe to. This kept the unpredictability of AI responses contained to the module that triggered them, without leaking into unrelated parts of the dashboard.

The frontend integrates with REST APIs for triggering and configuring AI workflows, and separately handles webhook-driven events that reflect asynchronous automation results arriving after the fact. A shared, typed API client layer handled the request/response boundary consistently across every AI feature module.

Authentication and authorization followed the same JWT-based pattern used across the organization's platforms, with role claims gating access to administration and analytics views versus standard workflow usage.

• JWT-based authentication and authorization gating access to workflow configuration and administration views
• Role claims distinguish standard workflow users from platform administrators

• Well-structured frontend architecture makes AI integrations easier to maintain — treating unpredictable AI output as just another data shape to defend against, rather than a special case, kept the UI resilient.
• Sharing one real-time event pipeline across AI features avoided a proliferation of one-off polling logic per module.