Form-aware extraction
Schema per form type, not a generic blob.
LogisticsAI Automation
Document AI for logistics ops
An end-to-end document automation pipeline that parses bills of lading, customs forms, and invoices, then writes structured data straight into the TMS.
- Client
- Northbound Freight
- Timeline
- 9 weeks
- Team
- 3 engineers
- Year
- 2024
12k+Docs / month
99.4%Field accuracy
8xFaster than ops
$0Re-keyed entries
What we walked into.
Operators were re-keying every shipment by hand into the TMS. Nights and weekends were a backlog of paperwork, and the error rate was creeping up.
- Forms came in 17 different layouts across carriers
- OCR was 'good enough' but not for billing-grade data
- TMS integration required extreme care on writes
Our approach, in four moves.
Tight feedback loops, weekly demos, code you fully own.
01→
Form taxonomy
Catalogued every form type, then built a labeled set of 4,000 documents to evaluate against.
02→
LLM + classifier
Two-stage pipeline: a fast classifier routes the doc, then a focused LLM extracts the schema.
03→
Human-in-the-loop
Confidence-based queue so ops review only the ~3% the model is uncertain about.
04
TMS write path
Idempotent, auditable writes with a one-click rollback if anything looks off downstream.
Gallery
A look at the build.
What we built
Every surface, considered.
Confidence routing
Auto-pass, review, or reject — never silent.
Audit trail
Every field has a source and a timestamp.
TMS integration
Idempotent writes with rollback.
Retraining loop
Reviewer fixes feed back into the model weekly.
Ops dashboard
Throughput, SLA, and accuracy at a glance.
Tech stack
What's under the hood.
AI
LangChainOpenAICustom classifier
Backend
PythonFastAPIPostgres
Infra
AWS LambdaS3EventBridge
Integration
TMS APIWebhooksSFTP
Outcomes
The numbers we are proud of.
Across the top 12 form types.
From 4 minutes to 30 seconds per doc.
Auto-processed with no overtime.
Reallocated from data entry to growth.
“Our ops team used to dread Mondays. Now the queue is empty by 9am. The pipeline just works, and when it doesn't, we know why.”
TE
Tom EriksonHead of Operations · Northbound Freight
More work
Keep exploring.
Healthcare & Medical Technology
Medipulse2026
MediPulse – AI-Powered Digital Healthcare Platform
Delivered measurable business impact.
Mobile ApplicationAdmin PanelAI LLM
View case study10–14 Weeks
FinTech
TradingSmart2026
Paper Trading & Strategy Simulation Platform
Delivered measurable business impact.
Website DevelopmentAI Automation
View case study32 weeks
Healthcare & Medical Technology
Monitraq2026
Monitraq – AI-Powered Remote Patient Monitoring & Healthcare Ecosystem
Delivered measurable business impact.
Mobile AppWeb App
View case study30 weeks
Let's build
Have a project that deserves the same treatment?
Tell us where you are, and we'll send a same-day plan with a clear scope, timeline, and price. No pitch decks.