Moka Eva
Eva is an AI recruiting platform I designed from 0-1 in 4 months. I built the calibration system and interaction patterns that enable true human-AI collaboration, resulting in 54% day-1 retention and an NPS of 29.
AI Hiring Agent
SaaS
Company
Moka
Moka is a leading HR SaaS platform serving enterprises like Nvdia, Tesla, and DJI.
To differentiate in a crowded market, they built Eva AI, a lightweight, AI-first recruiting tool. As a founding product designer, I delivered the 0-1 design, shipping the product in 4 months.
USER
The Recruiter's Challenge
Our core user is the recruiter, connecting candidates and hiring managers throughout hiring.
User interviews revealed key struggles: translating vague needs into clear criteria, identifying the right fit with confidence, and managing slow manual evaluation.
Landscape Research
Market Gaps & Opportunity
I analyzed five major AI recruiting tools to contextualize these pain points.
Most platforms automate sourcing and screening well, but lack dynamic criteria calibration beyond basic keyword matching. Feedback loops remain simplistic, limiting model improvement. While the industry shifts toward human-AI synergy, most tools lack seamless recruiter-in-the-loop collaboration.
These gaps defined Eva's opportunity.
INSIGHT
The Real Nature of Hiring Criteria
design goals
What Needed to Solve
Goal #1 Overview
How might we help recruiters and AI discover criteria together?
Step 1
Setting Up the Hiring Brief
Recruiters start by entering a job description. AI extracts requirements and generates a structured hiring brief. This raised two design questions: how should recruiters refine AI output, and what format works best?
The final solution is an inline AI guidance. It lets recruiters write, while the magic wand gives support directly in the flow, so they get guidance without losing control.
Before this, I tried versions like example prompts and a guided checklist, but they were either too vague or too rigid.
For the JD format, I explored tabs, forms, and an editable document.
Tabs split information across screens, and forms felt too rigid for nuanced hiring criteria, so the editable document worked best.
Step 2
Training AI Through Decisions
After processing the JD, the AI shows candidates for review, and each accept or reject helps it learn the recruiter’s criteria.
I tested two ways to present candidates. Showing all 10 at once felt overwhelming, so I chose one at a time to support deeper review.
After several rejections, the AI asks why, which turns those decisions into better feedback and training data.
Step 3
Making Feedback Feel Natural
After calibration, recruiters review the full candidate list and continue refining the results.
The final design uses AI-initiated suggestions with chat, so the AI can ask targeted questions and make feedback feel natural and easy.
Before that, I explored other approaches like yes/no feedback, magic comments, and drag-and-drop cards, but they either felt too simple, unclear, or clunky.
Goal #1 Learning
How Calibration Benefits Both Sides
The calibration system works on two levels.
For AI: it learns user intent through feedback, improving match rates from 30% to 75%.
For recruiters: it acts as scaffolding, letting them discover criteria by reviewing real candidates instead of defining perfect requirements upfront. The result: more accurate recommendations and stronger matches.
Goal #2 Overview
How might we enable confident decisions without overwhelming recruiters?
Exploration #1
Split-Screen for Efficiency
Exploration #2
Meaningful Information Over Metrics
With layout decided, the next question: how should candidate information be presented?
I explored three approaches: five-star ratings felt meaningless: users wanted to understand why someone fits, not just a score. Expanded detail views were too dense and slowed evaluation. The solution: qualification match highlights paired with AI-generated summaries explaining the reasoning behind each match. This cut evaluation time from 10 minutes to 2-3 minutes.
IMPACT
Shipped in 4 months & adopted by 43% of users
Eva shipped in 4 months. The calibration system closed the gap between recruiter intent and AI results: users found strong matches within 2-3 rounds instead of rejecting most candidates. Information design cut evaluation time from 10 minutes to 2-3 minutes.
Final results: 43% adoption rate, 54% day-1 retention, NPS of 29.
REFLECTIONS
Through designing Eva, I synthesized four AI design learnings
Complex information works better when it is revealed step by step.
Users trust AI more when clearly explains why it made a recommendation.
Even with automation, users still want control over the final decision.
A strong AI system should improve through user feedback while also helping users better understand their own preferences.
NEXT STEPS
What's Next: Bias & Automation
AI in recruitment is highly sensitive: bias detection and mitigation need stronger mechanisms. For instance, if recommendations skew toward one demographic, the system should proactively alert users.
AI-powered outreach is another area for exploration.
Reflections
What I've learned
Impact
Shipped in 4 months & adopted by 43% of users
Eva shipped in 4 months. The calibration system closed the gap between recruiter intent and AI results: users found strong matches within 2-3 rounds instead of rejecting most candidates. Information design cut evaluation time from 10 minutes to 2-3 minutes.
Key metrics: 43% adoption rate, 64% day-1 retention, NPS of 30.
Next Step
Bias handling & AI Outreach
AI in recruitment is highly sensitive, and clearer mechanisms for bias detection and mitigation are needed. For instance, if the AI’s recommendations skew heavily toward male candidates, the system should proactively alert the user.
AI Outreach remains an open area for further exploration.
