Work / Dispatch AI
Conard Logistics
A truck map that reads its own email
Every morning, carriers send Conard Logistics' dispatch team lists of available trucks: tables, free text, screenshots, even phone photos. We rebuilt the heart of the pipeline that watches that mailbox: a new AI extractor that reads every email, finds every truck and puts it on the live map dispatchers use all day, wall TVs included.
The challenge
What needed solving
Image-only emails were invisible to the old pipeline, so entire carriers never made it onto the map. And dispatch had been burned before by silent data loss. A missing truck is a missed load.
The fix had to be accurate enough to trust, and loud when anything went wrong.
Built with
- Claude (vision)
- Next.js
- Node.js
- Microsoft Graph
- MySQL on AWS RDS
- WebSockets
- Mapbox
What we built
- Vision-first extractionText, tables and images all go through the same Claude-based extractor, which returns structured trucks: carrier, location, availability date and equipment.
- Proven before it shippedWe hand-verified real carrier emails, including a 36-truck list, into a labelled evaluation set, and required a perfect score on every carrier before switching over.
- Dispatch rules, encodedRegion names like "FL Central" pin to a sensible city while keeping the original wording, and lanes count as trucks the way dispatchers count them.
- One-switch rollbackA single setting flips back to the previous extractor, which made going live a low-risk afternoon instead of a leap of faith.
Interface shown is an illustration with sample data. The production system holds client data and stays private.
Next project · Conard TMS 2.0
Rebuilding a trucking company's TMS without stopping the trucks
Got a problem like this one?
Tell us what's slowing your team down. We'll tell you honestly whether software can fix it.
Start a conversation →