Jas: AI interview builder for talent teams

Jas enabled customers to create tailored interviews in under an hour, reducing operational bottlenecks and supporting scalable customer growth

Before Jas, customers had to rely on our internal teams to define role requirements, identify the right competencies, and manually build interview. This process would take weeks.

With Jas, customers can now generate tailored interviews for the roles they’re hiring for less than an hour, reducing internal operational bottlenecks while improving speed, flexibility, and scalability as our customer base grew.

Customers simply paste in a job description, and Jas identifies the key competencies to assess and generates tailored interview questions grounded in our science.

I turned early proof-of-concept designs into a refined, self serve experience through user observation and feedback.

In March 2025, our engineering team prototyped a version using Claude, and I translated it into our design system for a fast, practical release.

During the beta phase, I joined customer calls with Customer Success and our PM to observe how customers used the product. They surfaced valuable insights into key pain points and confusion.

By March 2026, I synthesised these insights to redesign Jas into a more refined, self-serve experience.

I prototyped it in Cursor and demoed it to internal stakeholders (Engineers, product managers, leadership and customer success) for feedback. Through this feedback, I worked closely with product and engineering to refine the experience, weighing through priorities and feasibility.

After two rounds of feedback, I delivered a self-serve experience with the goal to drive the product toward a fully self-serve experience ready for customer release.

Jas should balance speed with the scientific rigour needed for accurate candidate assessment.

Moving forward, I have developed product principles that Jas should be grounded around to measure success.

Hiring teams enter the job description, Jas suggests competencies to assess candidates against, and then generates interview questions

Before the interview is generated, two steps matter most:

• The job description
• Competency selection.

A strong job description helps Jas evaluate candidates more accurately, while selecting the right competencies directly shapes the quality of candidates recommended to talent teams.

Hiring teams would sometimes input incomplete or poorly optimised job descriptions, reducing the quality of assessments generated.

The experience begins with users providing the job description for the role they are hiring for. Inputting a poor job description means a sacrifice in the quality of assessments.

New designs:

The new experience analyses the job description at the start, evaluates it against a set of best practice criteria, and generates targeted multiple-choice questions to fill in any missing context.

This gives Jas a much clearer, structured understanding of the role, enabling more reliable interview outputs, reducing the risk of inaccurate evaluation, and increasing hiring teams’ confidence in the assessments

Hiring teams require a lot of hand holding in the competencies selection process.

What I observed from customers:

• Sessions often stretched for hours, with repeated clarification on what each competency meant and how to select the right ones for the role.
• They would often override suggested competencies. This is behaviour we want to discourage. Their judgement might compromise the validity of the assessments generated. Guardrails around the self serve experience are essential.

New designs:

The hypothesis was that if we add in an AI chat assistant on the side, hiring teams will feel more empowered in their decisions during the process.

This would allow users to ask questions, refine competencies, and shape assessments around their ideal candidate persona for the role.

The goal was to reliance on expert support while helping maintain the quality and validity of assessments.

Competencies were difficult to visualise when presented separately

It can be difficult for users to visualise how a candidate may perform in the role based solely on a list of selected competencies. Translating these competencies into a description makes the assessment criteria more tangible to evaluate.

New designs:

Through back and fourth discussion with engineers, I introduced a candidate persona summary on the competencies page that describes the ideal candidate persona in plain language.

The hypothesis was that by making the persona more visible and understandable, users could make more informed decisions about which competencies to adjust.

Combined with the AI agent, users could simply tell Jas what they wanted. For example, "I want the ideal candidate to be more team-player focused", and Jas would adjust the competencies accordingly.

To make the experience truly end-to-end, hiring teams can start linking the interviews to their jobs so it’s ready to go

Jobs are often stored in hiring team’s ATS like Workday, SuccessFactors.

Thanks to integrations, hiring teams can link it to their jobs immediately after creation.

A visual language that strikes a balance between AI innovation and warmth.

Sapia.ai is built on AI, but our brand is grounded in humanity, ethics, and fairness.

I have tightened the visual language across the product, using Sapia.ai’s brand pink and purple, complemented by subtle gradients and thoughtful motion.

The result is a branding that feels modern, while remaining approachable and trustworthy.

Because this project is still ongoing, I was unable to obtain metrics for its success.

However these are metrics I propose to track for release:

• Time reduced to finish the interview creation process. Right now it’s sitting at ~35 minutes. The aim is to reduce to less than 20 minutes.
• AI chat adoption, questions asked and actions taken. By measuring this tells us the trust level from customers. If changes are frequent, this signals that something needs to be done to increase trust.
• Time spent per step: If they’re spending a significant time on a step, then that might mean we would need to do more refinements.
• Quality of hire and interviews produced: Continuously checking in with customers and our People Science team to assess interview quality and hiring outcomes.