AI-automated screening platform designed by humans for humans.
role
Product Designer
client
Banao Technologies
status
2025–26 · SHIPPED
timeframe
Ongoing · Team of 4
UX
RESEARCH
RTL
Most of this work is under NDA. What follows is the reasoning, and one screen.
Problem
An AI screening system produced more evidence about a candidate than anyone could read, and the product around it had been built screen by screen — no navigation, no overview, no way to do anything in bulk.
Solution
I worked across the HR side of the product: how someone moves through it, sets up a role, brings candidates in, and reads what comes out. Five internal HR interviews shaped most of it.
What I worked on
01
Getting around the product
Navigation
There was no navigation component. To move between sections, people went backwards — clicking back repeatedly until they landed somewhere they recognised, then forwards into where they actually wanted to be.
I added a persistent left navigation. Every section became reachable from anywhere, which meant the back button stopped being a navigation tool and went back to doing what it’s for.
Role setup
Creating a role ran through a long sequence of steps, and each one ended in a loading screen. The cost wasn’t the number of inputs — it was that every step made you wait before you could give the next one.
I grouped related inputs onto the same step. Fewer steps meant fewer waits, and questions that belonged together got asked together instead of being split across screens that had nothing to do with each other.
A place to land
There was no dashboard. Signing in put you straight into a list of jobs, which told you what existed but nothing about what was happening.
I introduced one, built around the few counts that answer the first question anyone has when they open the product: how many roles are open, how many interviews are scheduled, how many have finished. Not analytics — a status check.
Adding candidates
Candidates could only be added one at a time. For a role with a hundred applicants, that’s a hundred repetitions of the same form.
I worked with the engineering team on bulk upload by CSV, and a second path where uploading a resume lets the system extract the details and create the candidate from it.
Two paths, because HR teams arrive with different things. Some have a spreadsheet already. Others have a folder of resumes and nothing structured. Supporting only the first assumes a tidiness that isn’t there.
02
Reading the result
Two readers, one page
At the end of the process sits a report, and two different people read it: an HR generalist making the decision, and a technical reviewer checking the reasoning behind it. They need opposite things from the same page.
I put the system’s analysis above the score. A number invites a threshold — above the line or below it, done reading — so leading with the reasoning means the person deciding encounters why before how much.
Strengths and gaps are stated plainly rather than as a verdict. A verdict tells you what the system concluded; strengths and gaps give you something to weigh it against.
The deepest evaluation detail moved out of the primary read but stayed one action away. The generalist isn’t asked to interpret something they lack context for, and the technical reviewer still finds what they came for.
The constraint that shaped it
Partway through, the product’s position changed. What the system produced had been final and not overridable. It became a recommendation, with a person required to commit to it.
That raised the stakes on the hierarchy. If a human is accountable for the outcome, what they read has to make disagreeing possible. Leading with a number and burying the reasoning would have made the human step ceremonial.
03
A second language
Right to left
The platform expanded into an Arabic-speaking market. I worked with another designer on the right-to-left version of the full product, which deployed.
The hierarchy held — direction and layout changed, the order of the argument didn’t. Translation needed the most care. A literal translation can shift meaning, so wording was reviewed rather than swapped.