HPC work-order intelligence, dispatch routing, and automation across admin + technician personas.

Built for HackUTD’s NMC² (HPC Computing) challenge, this Next.js app is a full-ops cockpit: admins bundle and parallelize provisioning, technicians get map-aware walk orders, comms are auto-parsed into tickets, and inventory plus AI assistants close the loop for high-density NVIDIA fabrics.

• Metric grid, ticket board (known vs ambiguous signals), quick automations, cabling visualization, and predictive inventory sit in one hero screen.
• TicketBoard shows two ticket types—playbooks vs ambiguous hypotheses—wired directly to curated data in lib/tickets/data.ts.
• QuickAutomation actions tap a rotating credential pool to simulate push-button fixes for “press a button, get creds” tasks.

• Admin/technician tabs share filters for severity, tags, and channels.
• Toggle “Grouped bundles” to run the BundlingEngine heuristic that weighs severity, category, location, priorities, time windows, and customer to find parallel-safe batches.
• mapTicketToEnhancedTask enriches tickets with derived metadata (category, estimated duration, inferred location) before bundling.
• Ticket rows adapt to the persona: admins see queue + SLA data, technicians see floor, distance, and personal health cautions sourced from contexts/profile-context.tsx:1.

• Bundled tickets drive generateTechnicianRoute, which injects inventory pulls, elevator penalties, and distance scoring.
• findFacilityPath runs a multi-floor Dijkstra search over blocked aisles, hot/cold lanes, and lift transfers.
• Floor selection, bundle picker, route panel, legends, and cell modal expose annotations, telemetry, and 3D rack overlays. Opening a cell launches the GLTF rack sandbox so ambiguous issues can show likely causes in-context.

• Field crews can pick any staged device (H100, NVSwitch, QSFP, cooling loop, etc.) and either fire predefined prompts or ask custom questions.
• Prompts round-trip through the Gemini API route to return safety-checked installation/troubleshooting guidance under 200 words.

• Gmail unread threads and Slack IMs are normalized into Fabric tickets via GPT-4.1 prompts.
• The generator screen lets ops trigger inbox syncs, inspect structured payloads, and cache results into localStorage, which downstream pages (tickets + floor plan) read automatically.

• MongoDB-backed inventory exposes depletion scoring, weekly burn, filters, multi-select facets, and search across categories/location shelves.
• GET /api/inventory calculates health, weeks remaining, and warnings on the fly so panels and dashboards stay consistent.

• A GLTF-based rack view is accessible from both the dedicated /model sandbox and the floor-plan modal, complete with CSS2D callouts.
• The profile page persists technician health data locally; those constraints surface in technician tables (e.g., caution badges for multi-floor climbs).

• Framework: Next.js App Router + React 19 with client components for interactive dashboards.
• Design system: Hand-authored Tailwind-esque utility classes in app/globals.css, glassmorphism cards via SectionCard.
• State: Lightweight React hooks + context (ProfileProvider) plus browser localStorage caching for synced tickets.
• Data + AI services:
  • MongoDB (inventory) via lib/mongodb.ts.
  • Gmail OAuth helper routes (app/api/gmail/*) to capture refresh tokens or raw emails.
  • Slack user-token fetches inside the server route.
  • OpenAI GPT-4.1 (ticket normalization).
  • Google Gemini Flash (device assistant).
• Pathfinding + heuristics: Custom bundling engine, inventory-aware technician routing, and Dijkstra floor-walk calculations under lib/floor-plan.
• 3D + visualization: Three.js scene (threescene.tsx) dynamically imported wherever needed so the bundle modal can show rack context without blocking SSR.

1. Ticket ingestion
   • Gmail unread threads → GET /api/tickets/from-email → OpenAI JSON mode prompt → normalized Fabric tickets saved client-side.
   • Slack IMs → GET /api/tickets/from-slack → OpenAI prompt ensuring enum compliance → merged with manual backlog.
2. Bundling + routing
   • Tickets mapped to enhanced tasks (severity → duration, derived categories) → BundlingEngine groups them → Floor plan builds technician bundles → Routing determines best walk order, factoring inventory stops and elevator costs.
3. Device assistant
   • UX posts device context + question to /api/field-ops/device-assistant → Gemini Flash returns actionable steps.
4. Automation drawer
   • “Easy tasks” buttons rotate through credentialPool credentials to simulate just-in-time account provisioning.
5. Inventory
   • MongoDB collection documents augmented with depletion heuristics and weeks-remaining math before being returned to the UI.

Store these in .env.local. Gmail refresh tokens can be generated via /api/gmail/auth → Google consent → /api/gmail/callback.

1. Install dependencies

   npm install

2. Configure environment
   • Copy .env.local.example (create one if needed) and fill the variables above.
   • Seed Mongo with sample documents that match the schema used in app/api/inventory/route.ts:1, or point to an existing collection.
3. Run the dev server

   npm run dev

4. Optional integrations
   • Hit /ticketgenerator to sync Gmail + Slack tickets (requires envs).
   • Visit /floor-plan after syncing to see technician bundles incorporate real tickets.
   • Open /model to verify the GLTF scene renders before embedding it via the floor-plan modal.

1. Real work-order data – Swap lib/tickets/tickets.json for live service tickets or hook up a queue via tRPC/GraphQL.
2. Routing accuracy – Feed the pathfinding engine real facility geometry or add live telemetry for congestion/blocked aisles.
3. Automation hooks – Replace the credential simulator with real workflow actions (Terraform, NetBox, runbooks).
4. Inventory analytics – Stream inventory usage deltas from sensors/ERP instead of static Mongo docs and surface predictive restock alerts in the hero panel.
5. Technician safety – Extend the profile context with wearables or fatigue signals so dispatching can avoid overworking certain crews.

Let me know what to clarify or if you want help wiring any of these next steps!