ai assessment creation.

DoSelect already had an assessments product backed by a large question library for both tech and non-tech roles. On paper, the system worked.

In reality, assessment creation depended heavily on a Content & Delivery (C&D) team, who collaborated with recruiters and hiring managers to manually curate tests.

This led to:

• Multiple back-and-forth cycles for every assessment
• High turnaround time
• Recruiter dependency on hiring managers for basic inputs
• Lost customers because the system couldn't scale with demand

Recruiters lacked the confidence and domain knowledge to create assessments on their own, making the experience slow, opaque, and operationally expensive.

This wasn't just a UX problem — it was a scalability and automation problem.

At its core, the challenge was to move DoSelect from a service-assisted workflow to a self-serve, AI-assisted experience — without sacrificing assessment quality.

We needed to answer a few hard questions:

• How do we reduce dependency on the Content & Delivery team?
• How do we help recruiters think through assessment creation without overwhelming them?
• Where should AI assist, and where should humans stay in control?
• How do we design trust into AI-generated content?

This project required deep collaboration across Product, Engineering, and Content teams. My role focused on UX strategy and system design, not just screens.

Design process highlights:

• ▸Kicked off with PMs to review existing research, workflows, and experience gaps
• ▸Identified scope boundaries and design risks early
• ▸Shared early decision frameworks with the Content & Delivery team and iterated frequently
• ▸Presented direction and mental models to leadership — not just UI
• ▸Finalized flows, documentation, and handoff
• ▸Desk-checked production builds
• ▸Reviewed post-launch usage and feedback

This project evolved in public, with continuous iteration rather than a single "big reveal."

Early exploration narrowed down to two possible directions: a structured, form-heavy flow, or a chatbot-based, conversational flow.

We chose the chatbot approach. Because recruiters don't think in rigid fields — they think in context, intent, and trade-offs. A conversational interface allowed us to:

• Capture inputs incrementally
• Reduce cognitive load
• Avoid the feeling of "filling yet another form"
• Feed richer context into the AI system

This decision became foundational to DoIQ's UX and AI strategy.

The final direction reduced complexity while increasing clarity. The entire flow was condensed into three structured steps:

• 1.Upload job description — used as a reference, not a dependency
• 2.Confirm role and experience — helps AI reasoning and difficulty mapping
• 3.AI-assisted skill mapping table — this became the heart of the experience

While refining this system, we identified a new assessment type: Adaptive Assessments.

• Each skill contains questions across difficulty levels
• Candidates start with easier questions
• AI dynamically adjusts difficulty based on responses

This enables one assessment per role, more accurate skill measurement, and strong applicability in Learning & Development use cases rather than recruitment.

This insight emerged directly from system-level UX thinking, not feature requests.

Client accounts — 20,539 total questions · MCQs: 19,941 · Coding: 598

Internal accounts — 8,350 total questions · MCQs: 7,955 · Coding: 395

Internal teams began creating assessments independently using DoIQ Assessor, validating the self-serve vision.

Clients reported:

• Less need for question verification
• Higher confidence in AI-generated assessments
• Faster turnaround without C&D dependency