Freight operations
AFX Freight TMS
A freight brokerage platform that turns a shipment request into a quote and gives operators a clear point to review incomplete cargo details.
From shipment request to booking
A shipper starts with an origin, a destination, equipment, and cargo details. The brokerage needs to price the shipment and decide whether it has enough information to proceed. Those two decisions can have different answers for the same request.
The system I built
I built the React shipper and carrier portals and the Express API behind them. During my work at Melton, the system spanned 18 PostgreSQL tables and 35+ REST endpoints. I worked with engineering staff and leadership to turn brokerage requirements into technical specifications and acceptance tests.
Inside the platform
Architecture explorer
Follow a freight request through five parts of the system.
Two roles see different work.
Shippers request quotes and follow shipments. Carriers work with the loads assigned to them. Each portal presents the actions its user can take.
Authorization belongs on the server. A hidden button alone cannot enforce a role.
Pricing and booking are separate decisions.
The API connects quote requests, shipment tracking, and role-based workflows. A shipment can have enough information for pricing while an operator still needs to confirm its weight.
The quote replay below holds the shipment and market input fixed and changes only the weight.
Shared records give each portal a consistent view.
The platform keeps shipment and account records in PostgreSQL. Those stored records support the different shipper, carrier, and operator workflows.
A stored event supports a tracking update. An unconfirmed event still needs evidence.
Application data grounds a broker briefing.
The application computes the activity figures, then requests a written summary. In a local replay, the sample day had eight quote requests against a seven-day average of four.
An empty model response raised a warning and produced no briefing in the replay.
Deployment is part of the workflow.
I used Git-based deployments for the frontend and backend and included health checks. End-to-end and edge-case tests exercised the business rules around releases.
A service can respond to a health check while still making the wrong booking decision. The policy tests exercise that behavior.
Interactive example
Replay a quote
This local replay uses a fictional Tulsa-to-Dallas shipment with a fixed market input. Switch the weight and watch the booking decision.
Keeping pricing and eligibility separate
The server calculates the price and evaluates cargo eligibility separately. That gives a shipper a useful estimate even when an operator still needs to clarify part of the request. The interface can show a calculated amount and a request for more information at the same time.
A one-input replay
I replayed a fictional Tulsa-to-Dallas dry van request for 12 pallets of machine parts at 18,000 lb. With a fixed market input, the pricing engine returned $2,300 and marked the request eligible. Removing only the weight produced the same $2,300 quote and sent the booking decision to operator review.
Checking an AI briefing
Application code calculates recent activity before asking the model to summarize it. In an offline replay, the sample day had eight quote requests against a seven-day average of four. A valid simulated response produced a briefing. An empty response raised a warning and produced no briefing. Judging the quality of the writing would require real model outputs.