Beacon AI Demo

Beacon AI is a real-time support operations demo for an imaginary health center.
It lets users trigger realistic customer/patient scenarios and watch how inquiries are triaged, resolved, or escalated in real time.

Tech Stack

• Frontend: Next.js (App Router), React, Tailwind CSS, shadcn/ui
• Backend / DB: Convex (queries, mutations, actions, workflows, realtime subscriptions)
• Auth: Clerk
• AI: @convex-dev/agent with Anthropic models
• Workflow orchestration: @convex-dev/workflow
• Package manager: pnpm

Project Structure (high level)

• app/ - Next.js routes and UI
  • app/dashboard/customer-view/ - trigger scenarios and custom prompts
  • app/dashboard/admin-view/ - live operational columns (Active, Needs Human Review, Escalated, Resolved)
• convex/ - backend logic and schema
  • convex/schema.ts - data model and indexes
  • convex/inquiries.ts - inquiry lifecycle (create, admin actions, status transitions)
  • convex/agentService.ts - shared agent setup and thread message query
  • convex/agents/ - specialist agents (triage, scheduling, sales, emergency, retention, order)
  • convex/workflows/inquiry.ts - end-to-end routing and resolution workflow
  • convex/mockData.ts - seed data for demo scenarios
• lib/ - shared frontend/backend helpers and scenario config

Running Locally

1. Install dependencies:

pnpm install

2. Create your environment file and add values:

cp .env.local.example .env.local

Then fill in your variables in .env.local (for example: Clerk keys, Convex URL/deploy key, Anthropic API key).

3. Start Convex dev (backend):

pnpm convex dev

4. In another terminal, start Next.js:

pnpm dev

5. Open http://localhost:3000

Notes

• Use the Customer view to trigger seeded scenarios or send a custom prompt.
• Use the Admin view to monitor inquiry flow and intervene when needed.
• Inquiries flagged for human review or escalated trigger Slack notifications.