Skip to content
Reading documents that arrive via email, WhatsApp, and web portals Manually copying data from those documents into an internal TMS or spreadsheetManually copying data from those documents into an external system of - Shipping Line, Customs Authority or Port TerminalGenerating the next document in the chain — by opening a template and filling it in by hand
What goods are you shipping and how to handled them ?Who is shipping these goods? (shipper)Where is it coming from (where did the carrier receive it, from where the carrier is shipping it)Where is it heading (where will the carrier unload it from the shipment, and where will they deliver itWho is it being shipped to? (consignee)It defines the title of goods i.e one who has the original copy of the BL at the moment owns the cargoIt controls the payment settlement between the shipper and the consigneeIt serves as proof that the shipper has shipped the cargo It serves as the prerequisite document for the customs process in both the origin and the destination countries. Just like your passport is a prerequisite document for getting a visa
Booking confirmations arrive across email and portal downloads with no unified inbox → checking two places for every single shipment, multiple times a dayReviewing the confirmation against the requested schedule is entirely manual → no system flags mismatches, so a wrong vessel or cut-off date is caught only if the executive spots it themselvesEvery field extracted from the booking confirmation is typed into the TMS by hand →one shipment at a time, creating a direct pipeline for data entry errorsNo automated trigger to notify the shipper once the booking is confirmed → the update is sent manually, often delayed when the executive is handling multiple shipments simultaneouslyWhen a revision is needed from the shipping line, the follow-up loop is tracked nowhere → it lives in the executive's memory or a personal notepad, and drops get discovered only when the shipper chasesDocuments arrive across email, WhatsApp DMs, and group chats with no central inbox → the executive has to actively monitor 3–4 channels simultaneously just to know when a document has landedEvery document is reviewed manually against compliance requirements, field by field → there is no system that flags a missing HS code, a mismatched consignee name, or an incorrect port of dischargeGenerating a BL draft means opening a template, manually copying fields from the SI, CI, and Packing List one by one → a process that takes 25–40 minutes per shipment and is done entirely from memory of what goes whereRevision loops have no structure → a shipper's correction arrives as a WhatsApp message or a PDF annotation, and the executive has to manually locate the right field, update the shipment record, regenerate the draft, and resend - with no audit trail of what changed and whenMerge and split decisions are executed manually with no system guardrails - if a booking is incorrectly merged or split, the error propagates into the BL and can only be caught downstream, often after the draft has already been shared with the shipperNo real-time view of which shipments have pending document actions, the manager finds out about a stuck shipment either when the executive mentions it or when the customer escalatesException handling is entirely reactive, discrepancies, missing documents, and revision loops are invisible until they cause a delay, leaving no room for proactive interventionTeam workload is impossible to measure, there is no way to know how many documents an executive processed today, how long each took, or who is overwhelmed versus underutilisedSLA tracking is manual and retrospective, cut-off deadlines, BL release timelines, and shipper response windows are tracked on shared sheets that are updated inconsistently and always lag behind realityWhen a mistake surfaces, a wrong field on a BL, an incorrect booking confirmation sent to a shipper — there is no audit trail to understand what happened, who updated what, and at which step the error was introduced
Scattered → documents arrive across email, WhatsApp, and portals with no unified intake layer, the same document can arrive on three channels and no one knows which is the latest versionManual → every field extracted from every document is typed by hand into the TMS, one shipment at a time, by a human who could be doing something that actually requires judgmentBlind → no system knows what document is expected next, when it was due, or that it hasn't arrived yet, exceptions are discovered by the customer before the ops teamError-prone → a misread field, a skipped column, a copy-paste from the wrong document, any of these silently propagates through the BL and surfaces only at customs or at the shipping lineUntracked → revisions arrive as WhatsApp messages, get applied manually, and disappear, there is no audit trail of what changed, who changed it, and which version was ultimately sentUnify intake → receive and recognise documents from any channel - email, WhatsApp, upload → into a single processing queue, eliminating channel-hoppingAutomate extraction → read every document and populate shipment fields directly, replacing manual data entry with structured, validated outputFlag exceptions proactively → detect missing fields, compliance gaps, and data conflicts before they become a shipper escalation or a customs rejectionTrigger the next document automatically → know that once a Shipping Instruction is processed and approved, a BL draft should be generated, without waiting for a human to rememberMaintain a complete audit trail → every document received, every field extracted, every revision applied, logged with timestamp and source, visible to the entire teamGive managers real-time pipeline visibility → which shipments are waiting on documents, which are stuck in revision loops, and which are at risk of missing cut-off, without a single WhatsApp check-in
Map the document lifecycle for a single FCL ocean shipment — from booking confirmation to BL release. What documents are generated at each stage, and who is responsible? Where does the highest volume of manual work happen in this lifecycle? If you could automate only one step with an AI agent, which would it be and why? How would you define 'success' for this agent in Week 1 vs. Month 3? What does good look like at each stage? What are the three biggest failure modes of an AI agent that reads and acts on documents? How would you design guardrails for each? A document arrives with a field that conflicts with what's already in the system — for example, the consignee name on the SI doesn't match the booking. What should the agent do?
Document the detailed document workflow as per what we discussed in White boading roundIdentify the top 5 friction points in the current document workflow with evidence or hypotheses Define the problem spae for three personas: Documentation Executive, Customer Service Executive and the ManagersMap Jobs To Be Done for each persona — not features, but outcomes they need
Define the agent's scope: which document types will it handle in V1? Describe the intake layer: how does the agent receive documents across email, WhatsApp, and upload? Design the extraction logic: what fields must be extracted per document type? What is mandatory vs optional? Define the update logic: which shipment fields get updated from which document fields? What is the mapping? Design the exception engine: what conditions trigger a human review flag vs. auto-resolution? Describe the generation logic: given document type X has been processed, what document Y does the agent generate? On what trigger? Document arrives in a language other than English A required field is missing or illegible in the source document The same document is received twice from two different channels The extracted data conflicts with an existing confirmed field in the TMS
V1: What is the smallest useful version of this agent? V2: What gets added to make it meaningfully better? What would you defer entirely to V3 and why?
Define what success and failure will look like to customer Define what success and failure will look like to us will loo internally to usWill you launch this as an addon or default offeringWill you think about launching this as an independent product offering which can integrate with any TMS / ERP or part of Suite of Apps with TLP platform onlyHow will you think about pricing this product and why ?
A simulated document intake (show how Booking Copy will be received and how next steps will work, Show how the SI will be received and from that BL will be generated) Extracted field display — show what the agent understood from the document Exception flagging — highlight at least one field that triggers a review A preview of the next document that would be generated
Format: Notion / Coda / Google Doc + prototype link PRD should be implementation-ready — specific, not vague AI tools are encouraged — but be ready to defend every decision in the debrief
Share
Explore
The Multi-Billion Dollar Paper Jam
Autonomous International Trade Documentation Engine
🌍 Industry Overview
Think about the last time you booked a complex international trip - flights, hotels, transfers, and a visa. You could have done it all yourself, or you could have called a travel agent who knows the options, has the relationships, gets better rates, and sorts things out when it goes wrong.
A freight forwarder (3PL provider) is exactly that, except instead of moving people, they move cargo. When Zara wants to ship 10,000 denim cloth rolls from Arvind Mills in Gujarat to Amsterdam, they call a freight forwarder who books the container, arranges the truck to the port, handles customs on both ends, and tracks every step. They are the travel agents, project managers, and subject matter experts of global trade.
While platforms like MakeMyTrip, Expedia, ‘s — of the world revolutionised the experience of how we book our travels and stays within a few minutes, and Amedius, Sabre’s of the world, gave the connected backend platforms to travel agents to give the same experience with more customised offerings. Most of the Freight Forwarders still run on disconnected legacy systems, WhatsApp groups, Excel sheets, and institutional memory, while an importer/exporter might wait for 2–3 days to get a shipping quote, in an era where you can book a flight to Tokyo in 90 seconds. That's the problem space you're walking into. You don't need to be a logistics expert. You need to be a rigorous product thinker who can find signal in a messy, complex domain.
🧐 Problem Overview
While your tickets, boarding pass, and now your check-in process became completely digital with digiyatra, International freight still moves on paper or, at best, static PDFs shared over WhatsApp and email. A single shipment generates 10–15 documents across its lifecycle: Booking Confirmation, Commercial Invoice, Packing List, Shipping Instructions, Bill of Lading, Arrival Notices, Customs Filings, Delivery Orders, and more.
Today, a documentation executive at a freight forwarder spends 60–70% of their day on three things:
The result: delays, errors, missed SLAs, and an ops team buried in copy-paste work.
📖 Key Jargons
Term
What it means
Shipper (Exporter)
The business that manufactures or owns the goods and is sending them out of the country. Think of them as the seller in the transaction — for example, a textile factory in Surat shipping fabric to a buyer in Italy
Consignee (Importer)
The business or individual on the receiving end — they're buying the goods and importing them into their country. In the same example, the Italian fashion brand waiting for that fabric at the other end is the consignee.
Shipping Line
The company that actually owns and operates the ships that carry containers carrying the cargo across oceans — think Maersk, MSC, or CMA CGM. They're the airlines of the sea. The freight forwarder books space on their ships the same way a travel agent books seats on a flight.
Booking Confirmation
Once a freight forwarder requests space on a vessel, the shipping line responds with a Booking Confirmation — a document that says "yes, we've reserved a slot for your cargo on this ship, sailing on this date, from this port to that port." Think of it exactly like a flight booking confirmation
Commercial Invoice
The official bill raised by the shipper to the consignee — it states what was sold, at what price, and in what quantity. Customs authorities on both ends use it to assess duties and verify the value of goods crossing their border.
Packing List
Exactly what it sounds like — a detailed breakdown of what's inside each box, carton, or pallet in the shipment, including dimensions and weights. It's the cargo's identity document and is used by customs, the shipping line, and the warehouse to verify what's actually been loaded.
Bill of Lading (BL)
Imagine you’re a cargo owner in the year 1750, shipping exotic spices from India. You hand over your exotic spices to the ship owner (carrier) who will carry them across the world to Europe. In return for receiving your spices, you are presented with a bill of lading by the company that is shipping it
WTF is a bill of lading? It is a glorified piece of paper stating
Why is a bill of lading so important?
Shipping Instruction (SI)
Instructions from the shipper to the freight forwarder on how to prepare the Bill of Lading
There are no rows in this table
🔬 Problem Statement
Documents are the lifeblood of freight operations. Yet every document that arrives — via email attachment, WhatsApp forward, or portal (importer/exporter’s system or shipping line’s digital platform) download → requires a human to read it, understand it, extract data from it, update the shipment, and then produce the next document in the chain.
A Shipping Instruction arrives as a PDF on WhatsApp. The Docs executive opens it, reads the 14 fields, manually types them into the TMS, then opens a BL draft template in Word, fills it in, saves it, exports it as PDF, and sends it back to the shipper for approval. Average time: 40–45 minutes per shipment to generate the document and then 15 emails on avg to get the corrections and revisions to get to the final BL
Current Workflow — The Manual Chain
💊 Personas Involved & Pain Points
Persona
Description
Pain Points
Customer Service Executive
The frontline operator who owns the shipment from booking to departure. They place bookings with shipping lines, chase confirmations, cross-check every detail against what the customer requested, and are the first point of contact when something goes wrong on the ground. Their day is a constant loop of portals, inboxes, and WhatsApp →manually stitching together information from carriers, shippers, and internal teams.
Documentation Executive
The backbone of compliance and paperwork in a freight forwarding operation. They live inside documents → collecting Shipping Instructions, Commercial Invoices, and Packing Lists from shippers, validating every field against regulatory requirements, and manually generating Bills of Lading from scratch. A single typo or missing field can hold up an entire shipment at customs, so precision is non-negotiable. Their work is high-stakes, high-volume, and almost entirely dependent on how fast people respond on WhatsApp.
Managers of the Team
Responsible for keeping 100+ active shipments moving simultaneously without a single source of truth. They spend most of their time firefighting — chasing team members for status updates, manually reviewing exception cases, and answering customer escalations that should never have reached them. They have no real-time visibility into where each shipment stands, which documents are pending, or which team member is overloaded. Their ability to manage performance is limited to end-of-day WhatsApp check-ins and whatever their team remembers to update on the tracker sheet.
There are no rows in this table
🧠 Why a Document Intelligence Agent is Needed
The current document workflow is:
A Document Intelligence Agent would:
🧠 The Assignment
You are tasked with designing a
Document Intelligence Agent - Intelligent Queue / Inbox an AI-native layer that reads documents from any channel - Classification & Extraction - Claissify the documents received → extracts structured data → standardises the data from the documents - Creates / Updates shipment records - Flags exceptions if any - Generates the next set of document automatically - Generates a document as per user’s request when they share a template
🎯 Part A - White Boarding Round (60-90 mins)
There are no right answers — we are evaluating how you structure ambiguous problems, what questions you ask, and how you prioritise.
Some sample questions for you to get a head start
Lets see how you can use AI as companion before we get to actual White boarding round to research about the domain. To give you a quick head start you can look at companies like -
🛠 Part B - Take Home Assignment (4-5 days)
This is your opportunity to go deep. We expect a working prototype — not just a PRD — because we want to see how you use AI tools to accelerate your thinking and communication.
Deliverable 1 — Product Requirements Document
🔬 Problem Space
🧠 Solution Design
:a Laterals and Rabit Holes
🕛 Phased Roadmap
🏆 Success Metrics
💰 GTM and Pricing
Deliverable 2 — Working Prototype
Use Claude, ChatGPT, Lovable, Bolt, or any AI-native prototyping tool to build a working prototype that demonstrates the core agent flow. A static Figma mockup is not sufficient.
Your prototype must demonstrate at minimum:
Submission format: Loom walkthrough (5–7 mins) + link to live prototype
Submission Guidelines
Sample Documents
Booking Confirmation Samples
DB_aabhdabicbad0x0390.pdf
121.9 KB
MUMEE3539600.pdf
207.3 KB
Global_Booking_Confirmation_EBKG10896138.pdf
152 KB
HL-18153331_VNVUT_BC_ORIGINAL.pdf
54.2 KB
Bill of Lading Samples
House_Bill_of_Lading_Draft_HFSE24000418.pdf
4.5 MB
MCOP0101_495982786.pdf
163.8 KB
253256642.pdf
1.1 MB
Adobe_Scan_22_May_2025.pdf
877.8 KB
Want to print your doc?
This is not the way.
This is not the way.
Try clicking the ··· in the right corner or using a keyboard shortcut (
CtrlP
) instead.