About the role
You’ll sit with the people who review documents and exceptions, then build a working version of one part of their day. The job combines product discovery with hands-on software development. A promising prototype should answer a question about the workflow as well as the model.
You’ll follow one operator task from the original evidence through the final decision, then build that slice of the product across the interface and API. The useful question is often whether the reviewer can spot a bad suggestion quickly enough to correct it.
Working at ENNPRO
ENNPRO is an applied machine learning consultancy. We work alongside client engineers and the people who use their systems. Depending on the engagement, a client might need a service they can operate, a benchmark they can reproduce or enough evidence to decide whether to release a model. Their team should be able to use the result after handover.
Your closest collaborators will be a product owner, a model engineer and the people who handle exceptions in the client’s workflow. You’ll see a reviewer open three tabs to verify a single value and have the opportunity to build the change that brings the relevant evidence into one place. You’ll also see where a seemingly convenient shortcut makes a mistake harder to notice.
You’ll make product and technical decisions within a bounded workflow. The product owner helps define the outcome. The model engineer helps you understand what the service can return. You own the interaction between the two, including loading, correction, undo and the handoff to the next person. A feature is ready for a trial when a reviewer can use it on a difficult case and the team can reconstruct what happened.
Responsibilities
- Observe a review session with the product owner. Record which information the operator checks, what they copy between screens and where they ask another person for help.
- Build a working interface for the task and connect it to a model or replayed response. Include slow responses, missing evidence and the point where a reviewer can override the suggestion.
- Implement the feedback path across the client and server. Preserve the original suggestion, the final decision and enough context to investigate a disagreement.
- Watch someone use the prototype on representative cases. Take one finding into the next iteration and explain how you’ll tell whether the change helped.
- Map the states of the interface before building the happy path. Decide what the user sees if evidence is incomplete, two people edit the same item or a suggestion arrives after a reviewer has already acted.
- Implement a thin path through the application, from the database or API to the control the reviewer uses. Work with the client engineer on the event format so a correction is attached to the right record and version.
- Make trial observations useful to the next iteration. Agree on the question a session should answer, watch how users work around the interface, and bring one concrete change back to the team with a way to evaluate it.
Day to day
A realistic prototype can begin with recorded responses. This lets you exercise an ambiguous answer repeatedly and test how the interface behaves during a delay. Switch to the live service once you know what contract you need. Keep the recorded examples as fixtures for the component and the endpoint so a future change can be checked against the same cases.
You’ll use browser tools, application code and small data queries in the same investigation. When an operator says a correction “didn’t stick,” follow it from the control to the saved record and back to the screen. Show the team where its understanding of the interaction diverged from what the system did.
Projects and first 90 days
A representative assignment
A reviewer sees a proposed invoice category and a confidence score. Build a version that places the relevant source line beside the suggestion, then let the reviewer correct it. Replay a wrong suggestion, a missing source and a delayed response.
The handoff includes a usable end-to-end flow, a clear shape for the correction event and findings from a trial. The next engineer should be able to reproduce what the reviewer saw.
During a trial, two reviewers might give the same suggestion different answers for sound reasons. Capture the evidence and workflow state available to each person. That gives the product owner a basis for deciding whether to revise the label definition, change the interaction or involve a domain specialist.
How your first engagement could unfold
First month. Observe a review session and map one task from intake through the final action. Collect examples that cover a routine case, a doubtful suggestion and a missing piece of evidence. Build a small interactive version that a reviewer can try using those examples.
Second month. Connect the interaction to a test endpoint and persistence layer. Work through keyboard navigation, failure feedback and the correction event. Ask an operator to use it on unfamiliar examples and watch where the interface makes them stop to ask you a question.
Third month. Run a bounded trial with the product owner. Review corrections alongside handling time and downstream rework. Make one iteration based on what people actually did, then leave the client engineers a working test and the reasoning behind the resulting interaction.
Qualifications
Core experience
- You’ve shipped a feature across a browser interface and an API, and can explain the states a user encounters while a request is in flight.
- You can build accessible controls, handle partial failures and design a data model that keeps an edit attributable to the right record.
- You can describe a product decision that changed after you watched someone interact with your work.
- You can reason about state on both sides of an API, including a request in flight, a stale browser view and an edit made in another session.
- You can make a small interaction understandable and usable by keyboard and explain how you test it with someone outside the engineering team.
Useful additional experience
- Experience with a browser framework such as React, a typed interface between frontend and backend, or visual regression tests.
- Work on annotation software, case review systems, operational dashboards or an interface where users correct a machine-generated suggestion.
- Familiarity with trial design, event instrumentation or evaluating the quality of feedback before using it as a training label.
How to apply
Choose one interface you helped build. Describe what the user was trying to do, the implementation you owned and one observation that led you to change it.
An annotated screen, a short description of a user session or a sequence of events can all illustrate your work. Explain what you personally implemented across the interface and service, and include one detail you would now design differently because of what you learned from users.
You can use an anonymized example for confidential work. The application form includes space for your account of the project, an optional résumé and an optional link to a work sample. It currently lets you prepare and review a draft in your browser.
Equal opportunity and accessibility
ENNPRO welcomes applicants of all backgrounds. We consider candidates on the skills and experience relevant to the role, including experience gained through different educational and career paths. If you need an accessible version of these materials or an adjustment to the application process, write to contact@ennpro.pro.
Start application