AI-powered Resume Screening
Surfacing top candidates from large applicant pools for recruiters
I led the design of AI-powered resume screening workflows for Moka's ATS across 2 product phases — from MVP launch with seed user testing through data-informed phase II iteration.
Time
Jul - Mar 2026
Client
MokaHR
My Role
Product designer
Overview
Background
MokaHR is an enterprise HR SaaS serving 2K+ mid-to-large clients. Three forces drove the decision to integrate AI into current hiring products.
Unlock usage-based revenue
Moka’s traditional per-seat pricing plateaus when client’s HR teams stop growing. Usage-based AI features open revenue that scales with hiring volume, not team size.
Leverage domain expertise
Platform giants like Lark are bundling HR tools. Moka's edge is specialized models trained on recruitment workflows — differentiation general-purpose AI can't replicate.
Accelerate hiring cycles
LLMs enable a shift from workflow automation to active AI assistance — systems that handle repetitive evaluation and coordination tasks to shorten the time recruiters spend getting from application review to interview.
AI Vision Plan
I partnered with PM to map AI opportunities across the full hiring journey. Screening was prioritized for Phase I — it accounts for 60–70% of recruiter time and concentrates the most pain points.
My Contribution
End-to-end Ownership
I owned the full screening journey — from candidate shortlist through individual evaluation — across two phases, from MVP scoping through post-launch iteration.
Discovery
Collaborated with data analytics to track user behavior. Conducted user interviews to synthesize user goals and pain points.
Project scoping
Partnered with PM on phasing, feature prioritization, and user stories. Translated research findings into design requirements and negotiated scope tradeoffs between phases.
Interaction design
Defined how recruiters interact with LLM-generated outputs — from match indicators and evidence citations to a context-adaptive AI chatbot.
Problem Framing
Problem
Recruiters spend ~3 days per opening on resume screening with traditional ATS tools.
Root Causes
- It’s hard to surface strong candidates with keyword search only
- It’s time-consuming to manually find evidence of candidate qualifications
Problem
Three jobs-to-be-done synthesized from research guided design scope.
Design Process
User Story 1 - Surface Top Match
As a recruiter, I need to see a recommended list of the best match candidates and explanation of requirements they met, so that I have clear assurance the recommended list is significantly better than the rest.
Design Challenge
How should the system determine and present "top match" in a way recruiters can trust?
Phase I Shipped
A 'Top match' pre-prompt gave recruiters a one-tap shortcut to surface candidates who met the most high-priority requirements."
What Phase I Seed Data Revealed
After 2 months of seed use, I analyzed actual shortlisting behavior and discovered:
- 85% of shortlisted candidates matched all must-have requirements and 2+ preferred qualifications — suggesting a consistent, repeatable pattern behind recruiter judgment.
- Recruiters treated some requirements as binary but needed match depth on others — a single "match / no match" system couldn't capture this.
These Patterns Pointed to a Structural Solution
Auto-generate tiers based on requirement coverage thresholds recruiters were already applying intuitively
Phase II Response
Pre-prompt of “top match”
Tiered candidate list
Structured tiers replaced the pre-prompt. The system auto-arranges candidates by requirement coverage and match depth, with plain-language explanations.
User Story 2 - Refine the Shortlist
As a recruiter, I need to narrow down to a focused shortlist with candidates strong in different areas, so that I have a balanced set to advance to screening calls.
Design Challenge
How should recruiters query and filter candidates when their criteria go beyond pre-defined requirements?
Phase I Shipped
A natural-language search bar preserved the legacy ATS pattern for custom filtering.
What Phase I Seed Data Revealed
Seed user search queries revealed consistent patterns — experience filters, company signals, background traits — templatable enough for shortcuts, but too varied for rigid filter sets.
Phase II Response
Natural language search bar
AI chatbot with pre-prompts
An AI chatbot replaced the search bar. Where search filters by keywords, the chatbot reasons across multiple requirements simultaneously — enabling holistic evaluation.
Pre-prompts informed by seed user queries give shortcuts to experienced recruiters and teach strategies to novice ones.
Phase II Response
Structured filters
Free-form prompting
On-demand list
The chatbot returns a candidate list as a reviewable package — a working document that scopes to the recruiter's current screening step.
User Story 3 - Evaluate Candidate Fit
As a recruiter, I need to evaluate candidate qualifications and how well they match to job requirements, so that I can quickly judge if they are worth advancing to interview.
Design Challenge
How should the system communicate candidate-requirement match at different stages of the screening process?
Early Prototype Testing Insights
Prototype testing with 3 recruiters revealed uncertainties about AI evaluation:
- Match percentages felt arbitrary — minor differences between candidates carried no meaningful signal without a visible scoring logic.
- Requirement-level reasoning at scan level was overwhelming — recruiter wanted speed, not depth
Phase II Response: Progressive disclosure of requirement match
1
quantity at list level
2
match level on demand
3
reasoning & evidence at individual level
- Scanning: AI summary communicates strengths, differentiators, and gaps at list level.
- Evaluating: Expanded card shows match levels per requirement, ordered by priority.
- Validating: Full AI reasoning maps resume lines to requirements — recruiters verify the system's judgment before deciding.
Candidate List
Candidate Profile
AI-assisted Resume Screening
Final Prototype
Impact
From screening to confirmation to scheduling, the automation features reduced manual workload and improved conversion across the hiring journey.
25%
reduction in time-to-interview**
45%
trial-to-paid conversion on AI recruitment tools**
30%
increase in AI trust rating*
* From early prototype usability testing with 3 recruiters
** Phase I MVP launch across screening, outreach, and scheduling features among 60 seed users within 3 weeks.
Let’s Connect!
© 2025 Yiqing Wang
AI-powered Resume Screening
Surfacing top candidates from large applicant pools for recruiters
I led the design of AI-powered resume screening workflows for Moka's ATS across 2 product phases — from MVP launch with seed user testing through data-informed phase II iteration.
Time
Jul - Mar 2026
Client
MokaHR
My Role
Product designer
Overview
Background
MokaHR is an enterprise HR SaaS serving 2K+ mid-to-large clients. Three forces drove the decision to integrate AI into current hiring products.
Unlock usage-based revenue
Moka’s traditional per-seat pricing plateaus when client’s HR teams stop growing. Usage-based AI features open revenue that scales with hiring volume, not team size.
Leverage domain expertise
Platform giants like Lark are bundling HR tools. Moka's edge is specialized models trained on recruitment workflows — differentiation general-purpose AI can't replicate.
Accelerate hiring cycles
LLMs enable a shift from workflow automation to active AI assistance — systems that handle repetitive evaluation and coordination tasks to shorten the time recruiters spend getting from application review to interview.
AI Vision Plan
I partnered with PM to map AI opportunities across the full hiring journey. Screening was prioritized for Phase I — it accounts for 60–70% of recruiter time and concentrates the most pain points.
My Contribution
End-to-end Ownership
I owned the full screening journey — from candidate shortlist through individual evaluation — across two phases, from MVP scoping through post-launch iteration.
Discovery
Collaborated with data analytics to track user behavior. Conducted user interviews to synthesize user goals and pain points.
Project scoping
Partnered with PM on phasing, feature prioritization, and user stories. Translated research findings into design requirements and negotiated scope tradeoffs between phases.
Interaction design
Defined how recruiters interact with LLM-generated outputs — from match indicators and evidence citations to a context-adaptive AI chatbot.
Problem Framing
Problem
Recruiters spend ~3 days per opening on resume screening with traditional ATS tools.
Root Causes
- It’s hard to surface strong candidates with keyword search only
- It’s time-consuming to manually find evidence of candidate qualifications
Problem
Three jobs-to-be-done synthesized from research guided design scope.
Design Process
User Story 1 - Surface Top Match
As a recruiter, I need to see a recommended list of the best match candidates and explanation of requirements they met, so that I have clear assurance the recommended list is significantly better than the rest.
Design Challenge
How should the system determine and present "top match" in a way recruiters can trust?
Phase I Shipped
A 'Top match' pre-prompt gave recruiters a one-tap shortcut to surface candidates who met the most high-priority requirements."
What Phase I Seed Data Revealed
After 2 months of seed use, I analyzed actual shortlisting behavior and discovered:
- 85% of shortlisted candidates matched all must-have requirements and 2+ preferred qualifications — suggesting a consistent, repeatable pattern behind recruiter judgment.
- Recruiters treated some requirements as binary but needed match depth on others — a single "match / no match" system couldn't capture this.
These Patterns Pointed to a Structural Solution
Auto-generate tiers based on requirement coverage thresholds recruiters were already applying intuitively
Phase II Response
Pre-prompt of “top match”
Tiered candidate list
Structured tiers replaced the pre-prompt. The system auto-arranges candidates by requirement coverage and match depth, with plain-language explanations.
User Story 2 - Refine the Shortlist
As a recruiter, I need to narrow down to a focused shortlist with candidates strong in different areas, so that I have a balanced set to advance to screening calls.
Design Challenge
How should recruiters query and filter candidates when their criteria go beyond pre-defined requirements?
Phase I Shipped
A natural-language search bar preserved the legacy ATS pattern for custom filtering.
What Phase I Seed Data Revealed
Seed user search queries revealed consistent patterns — experience filters, company signals, background traits — templatable enough for shortcuts, but too varied for rigid filter sets.
Phase II Response
Natural language search bar
AI chatbot with pre-prompts
An AI chatbot replaced the search bar. Where search filters by keywords, the chatbot reasons across multiple requirements simultaneously — enabling holistic evaluation.
Pre-prompts informed by seed user queries give shortcuts to experienced recruiters and teach strategies to novice ones.
Phase II Response
Structured filters
Free-form prompting
On-demand list
The chatbot returns a candidate list as a reviewable package — a working document that scopes to the recruiter's current screening step.
User Story 3 - Evaluate Candidate Fit
As a recruiter, I need to evaluate candidate qualifications and how well they match to job requirements, so that I can quickly judge if they are worth advancing to interview.
Design Challenge
How should the system communicate candidate-requirement match at different stages of the screening process?
Early Prototype Testing Insights
Prototype testing with 3 recruiters revealed uncertainties about AI evaluation:
- Match percentages felt arbitrary — minor differences between candidates carried no meaningful signal without a visible scoring logic.
- Requirement-level reasoning at scan level was overwhelming — recruiter wanted speed, not depth
Phase II Response: Progressive disclosure of requirement match
1
quantity at list level
2
match level on demand
3
reasoning & evidence at individual level
- Scanning: AI summary communicates strengths, differentiators, and gaps at list level.
- Evaluating: Expanded card shows match levels per requirement, ordered by priority.
- Validating: Full AI reasoning maps resume lines to requirements — recruiters verify the system's judgment before deciding.
Candidate List
Candidate Profile
AI-assisted Resume Screening
Final Prototype
Impact
From screening to confirmation to scheduling, the automation features reduced manual workload and improved conversion across the hiring journey.
30%
increase in AI trust rating*
25%
reduction in time-to-interview**
45%
trial-to-paid conversion on AI recruitment tools**
* From early prototype usability testing with 3 recruiters
** Phase I MVP launch across screening, outreach, and scheduling features among 60 seed users within 3 weeks.
Let’s Connect!
© 2026 Yiqing Wang
AI-powered Resume Screening
Surfacing top candidates from large applicant pools for recruiters
I led the design of AI-powered resume screening workflows for Moka's ATS across 2 product phases — from MVP launch with seed user testing through data-informed phase II iteration.
Time
Jul - Mar 2026
Client
MokaHR
My Role
Product designer
Overview
Background
MokaHR is an HR SaaS serving 2K+ enterprise clients. Three forces drove AI integration.
Unlock usage-based revenue
Moka’s traditional per-seat pricing plateaus when client’s HR teams stop growing. Usage-based AI features open revenue that scales with hiring volume, not team size.
Leverage domain expertise
Platform giants like Lark are bundling HR tools. Moka's edge is specialized models trained on recruitment workflows — differentiation general-purpose AI can't replicate.
Accelerate hiring cycles
LLMs enable a shift from workflow automation to active AI assistance — shortening the path from application to interview.
AI Vision Plan
I partnered with PM to map AI opportunities across the full hiring journey. Screening was prioritized for Phase I — it accounts for 60–70% of recruiter time and concentrates the most pain points.
My Contribution
End-to-end Ownership
I owned the full screening journey — from candidate shortlist through individual evaluation — across two phases, from MVP scoping through post-launch iteration.
Discovery
Collaborated with data analytics to track user behavior. Conducted user interviews to synthesize user goals and pain points.
Project scoping
Partnered with PM on phasing, feature prioritization, and user stories. Translated research findings into design requirements and negotiated scope tradeoffs between phases.
Interaction design
Defined how recruiters interact with LLM-generated outputs — from match indicators and evidence citations to a context-adaptive AI chatbot.
Problem Framing
Problem
Recruiters spend ~3 days per opening on resume screening with traditional ATS tools.
Root Causes
- It’s hard to surface strong candidates with keyword search only
- It’s time-consuming to manually find evidence of candidate qualifications
Jobs-to-be-done
Three jobs-to-be-done synthesized from research guided design scope.
Design Process
User Story 1 - Surface Top Match
As a recruiter, I need to see a recommended list of the best match candidates and explanation of requirements they met, so that I have clear assurance the recommended list is significantly better than the rest.
Design Challenge
How should the system determine and present "top match" in a way recruiters can trust?
Phase I Shipped
A 'Top match' pre-prompt gave recruiters a one-tap shortcut to surface candidates who met the most high-priority requirements."
What Phase I Seed Data Revealed
After 2 months of seed use, I analyzed actual shortlisting behavior and discovered:
- 85% of shortlisted candidates matched all must-have requirements and 2+ preferred qualifications — suggesting a consistent, repeatable pattern behind recruiter judgment.
- Recruiters treated some requirements as binary but needed match depth on others — a single "match / no match" system couldn't capture this.
These Patterns Pointed to a Structural Solution
Auto-generate tiers based on requirement coverage thresholds recruiters were already applying intuitively
Phase II Response
Pre-prompt of “top match”
Tiered candidate list
Structured tiers replaced the pre-prompt. The system auto-arranges candidates by requirement coverage and match depth, with plain-language explanations.
User Story 2 - Refine the Shortlist
As a recruiter, I need to narrow down to a focused shortlist with candidates strong in different areas, so that I have a balanced set to advance to screening calls.
Design Challenge
How should recruiters query and filter candidates when their criteria go beyond pre-defined requirements?
Phase I Shipped
A natural-language search bar preserved the legacy ATS pattern for custom filtering.
What Phase I Seed Data Revealed
Seed user search queries revealed consistent patterns — experience filters, company signals, background traits — templatable enough for shortcuts, but too varied for rigid filter sets.
Phase II Response
Natural language search bar
AI chatbot with pre-prompts
An AI chatbot replaced the search bar. Where search filters by keywords, the chatbot reasons across multiple requirements simultaneously — enabling holistic evaluation.
Pre-prompts informed by seed user queries give shortcuts to experienced recruiters and teach strategies to novice ones.
Phase II Response
Structured filters
Free-form prompting
On-demand list
The chatbot returns a candidate list as a reviewable package — a working document that scopes to the recruiter's current screening step.
User Story 3 - Evaluate Candidate Fit
As a recruiter, I need to evaluate candidate qualifications and how well they match to job requirements, so that I can quickly judge if they are worth advancing to interview.
Design Challenge
How should the system communicate candidate-requirement match at different stages of the screening process?
Early Prototype Testing Insights
Prototype testing with 3 recruiters revealed uncertainties about AI evaluation:
- Match percentages felt arbitrary — minor differences between candidates carried no meaningful signal without a visible scoring logic.
- Requirement-level reasoning at scan level was overwhelming — recruiter wanted speed, not depth
Phase II Response: Progressive disclosure of requirement match
1
quantity at list level
2
match level on demand
3
reasoning & evidence at individual level
- Scanning: AI summary communicates strengths, differentiators, and gaps at list level.
- Evaluating: Expanded card shows match levels per requirement, ordered by priority.
- Validating: Full AI reasoning maps resume lines to requirements — recruiters verify the system's judgment before deciding.
Candidate List
Candidate Profile
AI-assisted Resume Screening
Final Prototype
Impact
From screening to confirmation to scheduling, the automation features reduced manual workload and improved conversion across the hiring journey.
30%
increase in AI trust rating*
25%
reduction in time-to-interview**
45%
trial-to-paid conversion on AI recruitment tools**
* From early prototype usability testing with 3 recruiters
** Phase I MVP launch across screening, outreach, and scheduling features among 60 seed users within 3 weeks.
Let’s Connect!
© 2026 Yiqing Wang