04
TL;DROperators dug through admin screens for routine answers → OpenAI over the shipment database plus live WebSocket tracking → 3 AI features in production, production crashes down ~20%.
Live site- Role
- Full-Stack Developer
- Client
- Allship.ai · United States
- Year
- 2023—24
- Stack
- React, TypeScript, Node.js, OpenAI API
In one paragraph
OpenAI answers over live shipment data; real-time tracking; tested dispatch flows. Operators answer routine questions without digging through admin screens, and dispatch flows stopped breaking under live updates.
Context
Allship connects shippers and carriers across the US. I worked on both sides of the stack for a year — dispatch tooling, tracking and the marketing surface.
Problem
Operators kept asking the same questions about loads, carriers and statuses, and answers lived across several collections. The dispatch UI also crashed under real-time updates.
What I did
- 01
LLM over the domain database
Integrated the OpenAI API with the shipment database so 20+ recurring operator questions get context-aware answers inside the dispatcher tool.
- 02
Live tracking
Driver status and shipment positions pushed over WebSockets with geolocation APIs instead of polling.
- 03
Kanban for dispatch
A ticketing board for async and sync work between dispatch teams.
- 04
Tests on the money paths
Jest suites around booking, assignment and Stripe-based payment flows.
Results
3
AI-powered features shipped to production
85%
test coverage on core flows
−20%
production crashes
+30%
engagement after the marketing redesign
Stack
- React
- TypeScript
- Node.js
- OpenAI API
- MongoDB
- WebSockets
- Stripe
- Jest