The problem
A freight logistics operation was processing roughly 1,000 shipment requests a month, arriving from vendors in every format imaginable: emails, inconsistent PDFs, even handwritten forms.
Getting that information into the Transportation Management System meant staff manually reading each document, decoding the details, and typing them in by hand.
A team of six spent roughly half their day on this alone, and multi-stop loads made it worse: pairing related shipments relied on staff memory and manual cheat sheets, often holding up carrier booking for a full day while someone tracked down the matching load.
The build
Symplytics built an AI-driven intake pipeline that reads incoming shipment documents (typed or handwritten) and extracts the fields the TMS actually needs: customer, ship date, pickup reference, origin/destination, pallet count, weight, freight class.
But rather than pushing that data straight into the TMS, it lands in a purpose-built verification portal where staff review, correct if needed, and approve with a click.
The system also flags multi-stop patterns automatically, surfacing when a new shipment likely pairs with one already waiting, instead of leaving that to memory.
The result: the team shifts from typing data to reviewing it, from data entry to operations judgment, with a human still confirming every record before it hits the TMS.
Technologies Used
| Technology | What It's Used For |
|---|---|
| n8n | Orchestrates the entire workflow — connecting inbox, AI, storage, and the verification portal into one automated pipeline |
| Gmail API | Monitors the intake inbox, retrieves incoming shipment documents and attachments automatically |
| Google Gemini 2.5 Flash | Reads and interprets shipment documents — including handwritten forms — extracting the details needed for scheduling |
| LangChain Structured Output Parser | Enforces a consistent data structure on every extraction, so output is always clean and predictable regardless of the source document's format |
| Google Drive | Stores the original source documents, linked back to each record for full traceability |
| Google Sheets | Acts as the staging layer where extracted data lives until a human reviews and approves it |
| Custom Web Verification Portal | Gives staff a purpose-built screen to review, correct, and confirm AI-extracted data before anything reaches the TMS |
| Transportation Management System (TMS) Integration | Receives only verified, human-approved data — never raw AI output |
Build summary
This build starts by watching an intake inbox and picking up shipment documents the moment they arrive, regardless of format — typed, scanned, or handwritten. An AI model reads each document and pulls out exactly the fields a Transportation Management System needs: customer, ship date, origin, destination, pallet count, weight, and freight class. Rather than writing that data straight into the TMS, the workflow stages it in a review layer and generates a dedicated verification screen where staff can see the original document side-by-side with the extracted data, correct anything that's off, and approve it with a click. The system also watches for multi-stop patterns — flagging when a new shipment likely pairs with one already waiting — so that judgment call doesn't rely on memory or a spreadsheet cheat sheet. Every record keeps a traceable link back to its source document, so nothing is a black box. The whole pipeline is built on n8n rather than custom-coded software, which means it can adapt quickly if a vendor changes their document format or if the underlying AI model needs to be swapped out — without a rebuild.
The stack
Have a similar problem?
Bring your messiest process — we'll tell you honestly whether n8n is the right tool for it.