Nova — Delay Alert + Rescue Planner

Real-time linehaul delay visibility and automated rescue planning across the transportation network
Python
ReactJS
AWS Lambda
Kinesis
DynamoDB
SNS/SQS
CloudWatch
Step Functions

Problem Statement

Amazon's linehaul network moves thousands of trailers daily between fulfillment centers. When a vehicle encounters delays — weather, carrier issues, or operational disruptions — the downstream impact cascades across the network. Without centralized delay visibility, NOC (Network Operations Center) teams relied on fragmented carrier updates, phone calls, and manual spreadsheets to identify and respond to delays.

Nova provides a unified delay alert dashboard that ingests real-time vehicle telemetry and carrier status updates through Kinesis streams. When delays are detected, the system automatically categorizes them by reason code, calculates downstream EDD (Estimated Delivery Date) impact, and surfaces rescue recommendations. The Rescue Planner provides a 5-tab workflow (Home, Edit, Plan, Check, Follow Up) with role-based access for Internal NOC, Line-haul Associates, and ATS Surface teams.

System Architecture

Vehicle telemetry and carrier status updates flow into Amazon Kinesis Data Streams in real time. A Lambda consumer processes each event, correlating it with the planned transit schedule stored in DynamoDB. When a delay is detected (actual ETA exceeds planned yard time), the system generates an alert and publishes it to SNS for downstream consumers.

The Rescue Planner uses AWS Step Functions to orchestrate the multi-step rescue workflow. When a rescue is initiated, the state machine coordinates vehicle availability checks, carrier capacity queries, route optimization, and approval workflows — with each step backed by a dedicated Lambda function.

The React dashboard connects to API Gateway, which routes requests to Lambda handlers for delay queries, rescue CRUD operations, and LLM-powered analysis. CloudWatch Alarms monitor delay volume spikes and rescue SLA breaches.

Animated edges show the primary request path. Dashed edges show async/secondary flows. Drag to explore, scroll to zoom.

Data Model

Delay Alerts
ColumnTypeNotes
vehicleIdVARCHARPK — VRID
laneVARCHAROrigin→Destination (e.g. SEA1→PDX2)
destinationVARCHARDestination city
zoneENUMNorth, South, East, West
scacVARCHARCarrier SCAC code
reasonCodedByENUMWEATHER, AMAZON_TOC, CARRIER
delayHoursDECIMAL0 = on time, >0 = delayed
plannedYardTimeDATETIMEExpected yard arrival
etaDATETIMECurrent estimated arrival
alertTypeENUMLH (Linehaul) or MR (Middle-mile)
Rescue Records
ColumnTypeNotes
rescueIdVARCHARPK
originalVridVARCHARFK → delayed vehicle
rescueVridVARCHARReplacement vehicle assigned
laneVARCHARAffected lane
delayHoursDECIMALHours of delay triggering rescue
rescueStatusENUMPENDING, APPROVED, IN_TRANSIT, DELIVERED
retrievalTimeVARCHARPlanned retrieval window
reasonForDelayVARCHARRoot cause description
clientApprovalsARRAY[BOOL]3 stakeholder sign-offs

Rescue Workflow

1
Home

Dashboard overview of all active rescues, status distribution, and priority queue.

2
Edit

Update retrieval time and delay reason for the selected rescue record.

3
Plan

AI-powered rescue recommendations — carrier selection, route, timing, and cost analysis.

4
Check

Stakeholder approval workflow — 3 sign-offs required before rescue execution.

5
Follow Up

Post-rescue tracking — delivery confirmation, carrier performance, and incident closure.

LLM Enhancements

Delay Intelligence Brief

Auto-loads on the Delay Alert dashboard. Analyzes all active delays to identify the critical corridor, top delay reason, and provides a network-wide intelligence summary with recommended NOC actions.

Represents how I would build this system today with LLM capabilities
Rescue Recommendation Engine

Evaluates a delayed shipment against available rescue vehicles, transit times, and customer impact to recommend the optimal rescue plan — including carrier selection, timing, and cost trade-offs.

Represents how I would build this system today with LLM capabilities
Natural Language Delay Filter

Queries like "Show North zone vehicles delayed 6+ hours with high EDD impact" are interpreted by the LLM and translated into structured filters applied to the delay alerts table.

Represents how I would build this system today with LLM capabilities
Executive Summary Generator

Generates a concise executive-level summary of the current delay landscape, including delay distribution by reason code, highest-impact lanes, and recommended escalation actions.

Represents how I would build this system today with LLM capabilities

Key Metrics

150

Vehicles Monitored

50

Rescue Records

< 5 min

Alert Latency

3 Roles

Access Control