Shipment Documents, Read and Ingested Without Manual Entry
An AI document processor that reads shipment documents as they arrive and ingests the extracted data into the Exim Routes system — bookings, invoices, bills of lading, customs permits and certificates, without manual data entry.
Bookings, invoices and customs documents, read and keyed in by hand
Exim Routes receives bookings, invoices, bills of lading, customs permits and certificates from shippers, carriers and agents. Today they are read and keyed in by hand. The client needed a parser that picks documents up automatically, understands what each one is, pulls out the shipment fields and writes them into their system.
Want Your Own Shipment Documents Extracted, Not Keyed In by Hand?
We'll scope a pilot around your exact document types and volume — a grounded extraction pipeline with automated retries — with clear success metrics before any production commitment.
A pipeline that listens, extracts, grades and retries
A pipeline that listens to a mailbox and accepts uploads; converts PDFs (including scanned), Excel and CSV to text; uses an LLM to classify each document and a specialist extractor agent per type to pull structured JSON; grades every extraction and retries with feedback when quality is low; lets a person confirm or correct the document type before extraction; stores results, source files and an audit trail in PostgreSQL, with every LLM call traced.
Intake (Gmail push or UI upload)
Ingest store (dedupe by hash, queue)
Text extraction (PyMuPDF, OCR if scanned)
Classifier (an LLM picks the agent)
Human confirmation (optional)
Extractor agent (1 of 5, value plus source)
Grounding check (drops values not in the text)
Grader (scores 0 to 1)
Save to PostgreSQL — below 0.85 the extractor retries, up to 3 times
Supported documents
Booking confirmations
Carrier and forwarder bookings.
Shipment details
Bills of lading, sea waybills, waste forms, transport schedules, weighbridge tickets, delivery notes.
Commercial invoices
Standard commercial invoice fields.
Customs declarations
Permits, export declarations, certificates of origin.
Compliance certificates
Pre-shipment inspection, chemical analysis, marine insurance.
Reliability and security
- Secrets stay in git-ignored files
- Mailbox access is read-only with a query filter
- Extraction is grounded against source text
- Every event is in an audit log
- Ingestion is idempotent — dedupe by message ID / file hash
- Failed documents retry up to 3 times then stop as failed
Status and gaps to close before production
No API authentication yet; access-key based auth instead of an IAM role; a hard-coded local database password in the dev compose file; checkpoint state held in memory. Known limitations: Word files and standalone images are not parsed yet; the invoice extractor has no graded documents yet; extracted data is not yet pushed into the client's system.
Next: push shipments into the client's system through an agreed API; deploy to the cloud with an IAM role and secrets manager; add API authentication and role-based access; run the evaluation set in CI.
An engineering team that ships graded, auditable pipelines, not slide decks
AWS Advanced Tier Partner
Certified cloud and AI delivery, so your document pipeline runs on infrastructure built to AWS Well-Architected standards from day one.
Pilot-first engagement
We scope, build and validate against your real documents before you commit to a production rollout.
Grounded extraction with an audit trail
Extraction is grounded against source text and every event sits in an audit log, with every LLM call traced in PostgreSQL.
Graded, with automated retries
Every extraction is scored 0 to 1; anything below 0.85 retries with feedback up to 3 times before it's saved.
Trusted by forward-thinking teams
AWS-Certified Experts Delivering Cloud Excellence.
Explore AI Solutions in Other Industries
See how we've built agentic AI, document processing and computer-vision systems across fintech, HR tech, real estate and more.
View All AI SolutionsWant Shipment Documents Extracted,
Not Keyed In by Hand?
Tell us about your document volume and formats — we'll help you scope a grounded extraction pilot with a measurable accuracy baseline.