PowerAI Labs logo

PowerAI Labs

Project Type: Azure AI

  • TrafficIQ

    TrafficIQ

    TrafficIQ Dashboard
    TrafficIQ Dashboard — traffic overview with delivery route status.

    TrafficIQ — Supply Chain Transport Intelligence. A multi-agent AI platform that bridges enterprise warehouse management (D365 F&O), real-time traffic intelligence (Azure Maps), and IoT fleet telemetry (Azure IoT Hub). Six specialist AI agents — coordinated by a three-tier routing orchestrator — give dispatchers, fleet managers, and field service teams natural-language control over shipments, routes, fleet, IoT, work orders, and inventory.

    🏆 Winner — Best Use of Microsoft Foundry at the AI Dev Days Hackathon.

    Problem

    Supply chain teams lack real-time traffic intelligence in their logistics stack. Dispatchers check traffic manually, can’t dynamically re-optimize multi-stop routes, and have no automated way to update ETAs when incidents occur. Fleet vehicles run without IoT telemetry or predictive maintenance, technician scheduling is manual, SLAs live on spreadsheets, and inventory visibility is fragmented across warehouses. Result: missed delivery windows, unplanned downtime, inefficient routing, poor customer communication.

    Solution

    TrafficIQ deploys 6 specialist AI agents with domain-specific tools, coordinated by a three-tier routing orchestrator (sticky → keyword → LLM RouterAgent). Each agent runs on Azure AI Foundry (GPT-4.1) via the Assistants API with its own system prompt, tool set, and thread. Agents call real backend services — Azure Maps, D365 F&O MCP Server, Azure IoT Hub, Microsoft Dataverse, Power Automate — and hand off context as conversations move between domains.

    TrafficIQ Multi-Agent Architecture
    Multi-Agent Orchestrator — 6 specialist agents with a Microsoft-style golden-path architecture.

    Stack

    • Frontend: React 19, TypeScript 5.9, Vite 7, Fluent UI React 9, React Router 6, TanStack React Query 5
    • AI & Agents: Microsoft Foundry Agent Service (Assistants API), Azure AI Foundry GPT-4.1 + GPT-4.1-mini, custom Multi-Agent Orchestrator
    • Maps & Traffic: Azure Maps Control 3.5, Azure Maps Traffic / Weather / POI / Route Range / Snap-to-Roads APIs
    • ERP & IoT: Dynamics 365 F&O via MCP Server, Azure IoT Hub (device twins, geofences, driving alerts), Microsoft Dataverse (dual-write)
    • Platform: Microsoft Power Apps SDK, Azure Static Web Apps, Azure Functions API, Azure Key Vault, Power Automate, Web Push (VAPID)

    Status

    Active — hackathon-winning build. Production-deployable via a single PowerShell script that provisions Azure resources, imports the Dataverse solution, and deploys the SPA + Functions API.

    Multi-agent architecture

    The Chat Panel sends user input through TRAFI (the AI identity) into the MultiAgentOrchestrator. A three-tier routing pipeline — sticky context → keyword match → RouterAgent LLM classifier — selects one of 6 specialist agents. Each specialist runs on its own Foundry thread, calls real backend services, and hands off summarized context when conversation moves between domains.

    • Traffic Agent — 17 tools: geocoding, route planning, incidents, reroutes, truck routes, isochrone (reachable range), POI search, weather, snap-to-road, saved journeys
    • Supply Chain Agent — 11 tools: warehouse shipments, delivery schedules, multi-stop route optimization, ETA updates, inventory, KPIs, exception alerts, live tracking, proof of delivery
    • Fleet Agent — 7 tools: fleet status, driver performance, vehicle health, maintenance alerts, service history, reachable range, GPS snap-to-road
    • Operations Agent — 7 tools: work orders, technician availability, scheduling board, AI-optimized scheduling, returns approval
    • Field Service Agent — 7 tools: service requests, customer assets, SLAs, technician dispatch with skill + proximity matching, visit completion, field parts inventory, KPIs
    • IoT & Logistics Agent — 7 tools: batch shipment analysis, route alternatives, IoT device health, geofence alerts, driving behavior, connectivity alerts, IoT logistics overview

    Plus shared tools across all agents: navigate_to_page and show_input_form (Adaptive Card input). 56 function-calling tools total.

    TrafficIQ AI Chat Agent
    TRAFI — natural-language assistant with agent badges and real-time tool execution indicators.

    Key features

    • Shipment Dashboard & Tracking — real-time shipments with traffic delay indicators, priority badges, warehouse filtering. Live GPS tracking with an inline Azure Map mini-map, proof of delivery capture, and traffic-aware ETA updates.
    • Delivery Route Planner — multi-stop optimization with Azure Maps waypoint optimization (TSP-style, computeBestOrder=true), total distance / duration / traffic delay, weather and POI overlays, shareable optimized routes.
    • Fleet Management — interactive FleetMap with SVG truck markers, status-coloured badges, rich click popups (driver, speed, load, fuel), marker clustering, and driver performance analytics.
    • Azure Maps overlays & controls — unified control panel for road / satellite / night style, live traffic, weather, POI markers, clustering, truck-vs-car route comparison, and reachable-range (isochrone) polygons.
    • Predictive Maintenance — vehicle health scoring, prioritized alerts (critical / high / medium / low), service history, telemetry-driven proactive scheduling.
    • IoT Device Management — full device registry at /iot-devices: signal / battery / heartbeat / firmware, geofence alerts, driving-behavior alerts, vehicle assignment, active-shipment auto-population from assigned vehicle.
    • Field Service — service request management, SLA tracking, skill + proximity-based technician dispatch with ETA, visit completion with resolution notes and parts, field parts inventory, KPIs (first-fix rate, mean response time, SLA compliance).
    • Scheduling Board & Work Orders — visual scheduling grid, drag-and-drop, AI-optimized assignment, work order lifecycle, parts and inventory integration.
    • Inventory & Returns — multi-warehouse stock levels, low-stock / overstock alerts, return request workflows with reason-code analytics.
    • Analytics Dashboard — supply chain KPIs (on-time delivery, fleet utilization, SLA compliance, warehouse utilization, cost per km), clickable KPI cards, exception alerts that route to relevant pages, performance trend indicators.
    • TRAFI multi-agent assistant — natural language conversation with 56 tools, bilingual support (Danish / English), Adaptive Card forms, thread-based context handoff between agents, real-time tool execution indicators.
    • Commute Intelligence — personal journey planning, saved journey monitoring with morning alerts, historical route comparison, push and email notifications.
    TrafficIQ Delivery Planner
    Delivery Planner — multi-stop route optimization with traffic, weather, and POI overlays.
    TrafficIQ Fleet Management
    Fleet Management — real-time vehicle tracking with clustering and rich popups.
    TrafficIQ Analytics
    Analytics Dashboard — supply chain KPIs with clickable cards and exception alerts.

    Dataverse persistence (dual-write)

    All application data is persisted to Microsoft Dataverse with a localStorage cache for fast reads — 20+ Dataverse tables across shipments, warehouses, inventory, fleet, work orders, technicians, notifications, journeys, maintenance, returns, IoT devices, geofences, and KPIs. Settings dual-write covers 9 categories (demo mode, theme, map, D365, Dataverse, Dataverse MCP, email, IoT Hub, agents) into a tiq_appsettings key-value table. Three connection modes are supported: Power Platform SDK (embedded in Power Apps), Dataverse REST proxy (local dev with OAuth2 client credentials), and localStorage fallback for offline / demo.

    Hero technologies

    • Microsoft Foundry (Agent Service) — multi-agent system via Foundry REST API (agents, threads, messages, runs)
    • D365 F&O MCP Server — CRUD on warehouse shipments, inventory, delivery entities via data_find_entities / data_update_entities
    • Azure Maps — traffic incidents, route optimization, geocoding, weather, POI, truck routing, isochrone, GPS snap-to-road
    • Azure IoT Hub — GPS device management, geofence monitoring, driving behavior telemetry, fleet connectivity
    • Azure AI Foundry — GPT-4.1 + GPT-4.1-mini deployments for agent reasoning
    • Azure Key Vault — centralized secret management with Vite build-time injection
    • Azure Static Web Apps — SPA hosting with built-in Azure Functions API

    Key learnings

    • A three-tier router (sticky → keyword → LLM) gives the best balance of cost, latency, and accuracy for multi-agent orchestration — most turns never hit the LLM classifier.
    • Dual-write to Dataverse with a localStorage cache is the right pattern for Power Platform-embedded apps — fast UX, durable data, offline-tolerant.
    • Tool design dominates agent quality. 56 small, well-named tools with crisp schemas outperform a handful of “do-everything” mega-tools.
    • Bringing real enterprise systems (D365 F&O, Azure IoT Hub, Azure Maps) into a single conversational surface is now genuinely tractable with Foundry Agent Service.

    Repository

    GitHub repo: github.com/raghavmishrad365/TrafficIQ · Licensed under MIT.