A shared workspace for generating, comparing and checking shipping documents — with the source always within reach.
Shipping teams work across bills of lading, letters of indemnity, voyage instructions and financial statements. The same vessel, cargo and commercial details appear in several documents, often in different formats.
The challenge was to reduce repetitive work while keeping AI output easy to check, correct and trace back to its source.
I led discovery and all interface design, then worked with the product owner to prepare the designs and hand them over to development.
A coherent toolkit, complete task flows and detailed states for upload, preparation, generation, comparison, review, export and recovery.
Reuse source information and templates rather than asking operators to re-enter the same details.
Bring documents, extracted fields and differences into one workspace.
Let operators correct fields, reject matches, resolve missing information and choose when to export.
I led discovery across the document tasks. The final designs address three recurring needs: preparing messy inputs, checking uncertain results and avoiding a new interaction model for every tool.
Bundled PDFs, inconsistent formats and information pasted from email.
Missing fields, suggested document matches and totals that need verification.
Upload, process, review and export recur across the toolkit.
Bill of lading → Letter of indemnity
Recap → Voyage instructions
Bunker nomination → Instructions
Letters of indemnity
Agent & master statements of facts
Manifests, bills of lading & mate’s receipts
Hire statements
These provisional role profiles explain who the workflows serve and what each person needs to check. They are inferred from the designs, rather than presented as interview findings.
Works across bills of lading, letters of indemnity and manifests. Needs to resolve missing information without losing the original source.
Brings together recaps, ports, dates and bunker details. Needs to check that instructions reflect the current agreement.
Checks hire statements, balances and commercial terms. Needs precise discrepancies with enough context to judge them.
Needs a clear distinction between generated work and checked work, with an identifiable version to hand over.
The journey brings preparation, processing and source checking into one workspace. This map connects the workflow friction to the intended experience.
The proposed research moves from observation and interviews to prioritised requirements and task-based testing. It focuses on how operators judge whether a document is ready to use.
The 32-part UX dossier brings together the project brief, evidence register, stakeholders, role profiles, jobs to be done, current and future journeys, task analysis, information architecture, requirements and design traceability.
Ten document tasks, shared preparation and review patterns, source comparison, saved work and recovery states.
Role assumptions, the cost of switching documents, the meaning of match scores and whether the review steps catch consequential errors.
The proposed research includes contextual interviews and observation, recruitment across the four roles, synthesis, prioritisation and seven task-based usability scenarios. No participant findings or measured outcomes are claimed.
Read the full UX discovery board ↗Generation flows combine source documents with templates and company details. Comparison flows pair documents, highlight changes and keep the originals available for checking.
Split bundled files, confirm page ranges and resolve unsupported documents before processing.
Confirm or reject suggested matches using the actual vessel, cargo and document details.
Amend missing fields, inspect differences and verify totals before copying or exporting.
The detailed flows connect actual interface screens. Arrows start at the control the operator clicks; automatic transitions are shown separately. The comparison journey below follows upload, processing, review and output.
Start with two versions of a letter of indemnity.
Flow reconstructed from the designed states. Exceptions include rejected matches, stopped processing, missing information and unsaved changes.
The workspace changes to suit the task, while the header, progress and primary actions stay familiar.
Define what belongs on screen and where the next action sits.
Map the click, the response and the route back when something goes wrong.
Use hierarchy, status and proximity to make the review practical.
These wireframes were reconstructed from the designed journeys to explain the interaction logic. They are not original discovery artefacts.
Task cards make the expected input and output clear. Generate and compare are separate entry points, with favourites for frequently used tools.
For bundled PDFs, operators can divide pages manually, by a fixed count or with AI assistance. Thumbnails and page ranges make the preparation visible before the task runs.
Editable fields sit beside the original document. Missing information is called out, so the operator can resolve omissions before saving, copying or exporting.
Document pairs expose the details needed to confirm or reject a suggestion. A match score supports the decision; it does not replace checking the evidence.
Paired originals, labelled insertions and deletions, filters and change navigation help reviewers focus on relevant differences. An email draft can be copied after review.
The denser layout trades some reading space for continuity: the reviewer can inspect a change without leaving the comparison.
Manifest checks focus on cargo quantities and totals. Statements of facts focus on events, missing entries and timestamps. Hire statements focus on balances and discrepancies that need verification.
Voyage and bunker workflows bring together source text and structured operational details, so the final instructions can be reviewed before use.
A dark teal workspace frames the product. Purple and teal distinguish actions, while status colours are paired with labels so meaning does not depend on colour alone.
Roboto and Kanit support the product interface. Inter gives this case study the same editorial treatment as the portfolio.
Consistent headers, progress states, review actions and document panels keep navigation familiar as tasks change.
I worked with the product owner to hand the designs over to development. The handoff covered more than finished screens: it described how each task behaves while inputs, results and saved work change.
Supported files, bundled documents, page division and readiness checks.
Progress, stopping a task, failures and returning to a usable state.
Missing fields, rejected matches, amendments, saved work and unsaved-change exits.
Copying, downloading and exporting only after the operator has reviewed the result.
The interactive Figma prototype brings the document workflows, recovery states and a plain-English usage guide together. It uses sample data and simulated processing.
Open the working prototype ↗A North Star metric is the single measure that keeps design decisions focused on user value. For Shipster, the proposed measure is the percentage of started document tasks completed independently, reviewed against the source and exported without a critical error.
The target combines completion with accuracy. A quick export only counts if the document has been checked and contains no critical errors.
Shared preparation and source-adjacent review reduce switching and repeated entry.
Readiness checks and visible omissions catch problems before output.
A consistent workflow makes each tool easier to learn and complete.
All figures above are illustrative planning assumptions. Baselines must be established through equivalent task tests before improvements can be reported as outcomes.
Compare the current workflow with the redesigned workflow using equivalent documents, balanced task order and representative operators. Record every started task, completion, assistance, review, export, correction and critical error. Verify the output independently against the source.
Report task counts, sample size and uncertainty alongside the completion rate. Separate generation, comparison and matching tasks so a change in task mix does not create a false improvement. Capture median completion time and corrections, then ask operators to rate task ease from 1 (very difficult) to 7 (very easy).
Quality guardrail: critical errors must not increase. A critical error is an incorrect vessel, cargo, financial amount or operational instruction that makes the output unsafe or unusable. An unchecked export does not count as verified success.
Task-based testing should check whether operators can prepare a bundled PDF, resolve a missing field, reject an incorrect match and find a meaningful difference. Quality checks should compare results with the source. Operational measures should track completion time and correction effort.
Proposed validation measures; these are not presented as measured outcomes.
The core design decision was proximity: keep the source, the result and the next action close enough that checking becomes part of the workflow.