Artificial intelligence
How we solved it with Griver
Grupo Inversor Veracruzano, Mexico.
- Client
- Griver
- Country
- Mexico
- Sector
- Foreign trade
The problem
Mexican customs agencies review every pedimento and invoice by hand, a process that used to take a full day.
The solution
An AI OCR API for the RECO prevalidator and a self hosted language model; the data never leaves the client's infrastructure and is not stored.
Why it worked
What used to take a full day of review now takes 10 minutes with the RECO prevalidator.
In numbers
- 18 s
- of automatic processing per document (median)
- 11.546
- documents processed in one week
- 21.591
- pages read that same week
- 74%
- of documents arrive scanned, not digital
Live operations console, week of September 14 to 20, 2026.
The operations console
Real screens from the console in production, with amounts and names blurred. Hover to pause, or pick a stop on the rail.
Every page, measured
Pages per day, digital versus scanned documents, latency and errors, by environment and date range. The console only sees business metadata, never the documents.
On site
On the yard, before writing a line
We went to the port with Griver's team: to see where each document comes from, how a container moves and what a customs agent looks at when reviewing an entry. What gets built afterwards resembles what happens on the yard, not what you picture from an office.
Tools
- AWS GPU
vLLM
Qwen 2.5- Docker
- Cloudflare Tunnel
Facing a similar problem?
Tell us what you need. We review your case and reply by email.
Get a Quote
Complete the form and discover how we can help you achieve your growth goals with custom solutions.