Case 02 / 08 · AI truck dispatcher
Dispatch
One dispatcher now runs 15 trucks instead of 8, on a platform I built, sold and still keep secure.
The problem
A dispatcher spends the day on load boards, broker calls, driver texts and paperwork. Around eight trucks they run out of hours, and money leaks quietly: empty miles, waiting time nobody billed, invoices sent days late.
How it works
Six steps,one system.
- 01Scan load boards
- 02Rank loads per truck
- 03Counter-offer
- 04Driver check-ins
- 05Paperwork check
- 06Bill detention
Scanning · every 3 minutes
live6 of 15
It watches every board for every truck that is about to be empty.
Every few minutes the agent reads the load boards for every truck that is about to be empty.
Inside
What I built,layer by layer.
01
Five agents, one dispatcher
Separate agents for loads, broker offers, driver check-ins, paperwork and detention, each with one job and its own checks.
02
A person approves what matters
Nothing goes to a broker and no money moves until the dispatcher approves. Routine updates run on their own.
03
Reads the real paperwork
Rate confirmations, bills of lading and delivery photos are read, filed and compared, so a $200 mistake is caught before the invoice goes out.
Engineering decisions
01
Suggest, don't book
The agent ranks and drafts, the dispatcher clicks. Dispatchers trusted it in the first week because it never acted behind their back.
02
Every number has a source
Each ranking shows why: pay per mile, miles empty to pickup, driver hours left. If a dispatcher disagrees, they can see exactly where.
03
Hard limit on AI cost
A monthly ceiling the system cannot pass, and a backup AI provider if one goes down mid-shift.
Scale of the codebase
8 → 15
trucks per dispatcher
+11%
pay per mile
−18%
empty miles
Same day
invoicing, was 3 days
Built end to end and sold. I still run its servers, security and updates, and I designed and shipped the AI agents as a paid add-on after the sale.
Next case
Bastion →