03Logistics
Document automation for cross-border shipping
OCR + LLM pipeline that ingests bills of lading, flags discrepancies and clears paperwork without an analyst in the loop.
6×faster clearance
Client
Freight forwarder · 200+ shipments/week · India ↔ Gulf & SEA
Engagement
12 weeks · Advanced engagement
Stack
AWS Textract · Claude 3.5 Sonnet · Postgres · Customs API · Retool
01The Challenge
What we walked into
Each shipment generated 6-12 documents that an analyst had to read, cross-check and reconcile against the booking. A team of four was clearing roughly 50 shipments a day and consistently held up consignments by 36-48 hours.
02The Approach
How we tackled it
- 01Built an OCR + LLM pipeline that ingests every document the moment it lands in the shared inbox.
- 02Cross-checked weights, HS codes, values, and consignee details against the booking — flagged discrepancies in a Retool review queue.
- 03Auto-cleared the 70% of shipments with no discrepancies straight to the customs API.
03The Outcome
What changed
- Clearance time fell from ~38 hours to ~6 hours on average.
- Analyst team now reviews only the 30% of consignments with real issues — and handles 4× the volume.
- Discrepancy detection caught ₹14L of mis-declared value in the first quarter alone.
“
Our customers used to ask why a shipment was delayed. Now they ask how it cleared so fast. That's a much better problem to have.
Have a similar problem?
Let's build something worth using.
Tell us about the bottleneck. We'll come back with a free 30-minute call and a quick read on what's feasible.
Start a conversation→More work
View all →Retail
Inventory copilot for a 30-store chain
Replaced manual stock checks with a forecasting agent that cut out-of-stocks by 38% in one quarter.
−38%
stock-outs
Healthcare
Triage assistant for an outpatient network
Voice + chat intake that routes patients in under 90 seconds and books the right specialist on the first try.
90s
avg. triage