Myrzan Izimbetov

Product Designer / Design Engineer at BI Group

Past 2.5 years. 2M+ daily users. Every project shipped to production.

Real estate platform ranked #1 in sales volume across Middle East & CIS. I owned design end-to-end — from user research in sales offices to production-grade prototypes handed off via GitHub.

Search & Filters

+156% page depth · Bounce rate 25% → 16.8% · 679K filter interactions/month

75% of traffic was mobile, but the filter panel was buried under whitespace and unclear affordances. Users were leaving before finding a single listing.

I ran card-sorting sessions with 8 users in a sales office — gift cards, screen recordings, the full guerrilla setup. One finding drove the entire redesign: City, Price, and Rooms covered 87% of all filter use.

I surfaced those three as quick-filter chips above results and pushed everything else one tap deeper. Then restructured listing cards for faster vertical scanning on small screens.

The decision was deliberately reductive — cut options to accelerate decisions, not the other way around.

What I changed

• Simplified filter architecture around the three most decision-critical criteria
• Restructured listing cards for faster mobile scanning
• Moved secondary filters behind progressive disclosure

Impact

• 679K+ filter interactions in the first month
• Page depth up 156% — users exploring more listings per session
• Bounce rate dropped from 25% to 16.8%

Key takeaway

The biggest engagement gain came from removing options, not adding them.

Design System · WCAG 2.2 + Dark mode

−83% QA time per task · 50+ semantic tokens · WCAG 2.2 compliant · Dark mode shipped

The product had outgrown its components. Inconsistent states, undocumented patterns, and ad hoc colour decisions were causing repeated QA failures and slowing every feature release.

I audited the full component library, identified the 12 highest-friction patterns (ones causing the most QA tickets), and rebuilt them with semantic tokens, documented states, and WCAG 2.2 contrast compliance. Then extended the entire system to support dark mode — not as a skin toggle, but as a first-class token architecture.

I worked directly with frontend engineers to ensure every rebuilt component was easier to implement than the one it replaced. If a component was harder to use correctly, it wasn't done.

What I changed

• Audited and prioritised components by QA ticket volume
• Rebuilt 12 high-friction patterns with semantic tokens and documented states
• Designed a token architecture that supports light/dark as first-class modes
• Paired with engineers to validate implementation ergonomics

Impact

• QA time per component down 83%
• Eliminated an entire category of recurring visual inconsistency bugs
• Dark mode shipped product-wide without new QA overhead

Key takeaway

A design system earns its value when engineering ships faster with fewer defects — not when the Figma file looks organised.

AI Prototyping · BI Tinder · Swipe Discovery

Concept to stakeholder demo in 1 day · Built in production stack · Shipped via GitHub

Growth hypothesis: could apartment browsing convert better as a preference-based matching experience instead of a traditional list?

Most designers would spec this out and wait for a sprint slot. I built it — in our actual production stack using Claude Code, not a Figma prototype. Same API, same component library, same data. The prototype was functional enough to hand off via GitHub, not a throwaway demo.

Stakeholders saw a working product the same day the idea surfaced. The concept is now in active development.

The point wasn't that I used AI tooling. The point is that a design hypothesis that would normally block the roadmap for weeks was validated in hours without taking a single engineer off their current work.

What I did

• Reframed listing browsing as a swipe-based matching interaction
• Designed the complete interaction model and user flow
• Built the prototype in the team's real tech stack (not a throwaway)
• Handed off production-ready code via GitHub

Impact

• Concept to stakeholder decision in 1 day
• Zero engineering time consumed from the team's roadmap
• Prototype now in active development

Key takeaway

AI-assisted prototyping is most valuable when it lets you validate hypotheses at production fidelity without blocking anyone else's work.

Beeline: UX research bundle in E-com->Beeline — From Design Porridge to Systematic Harmony->

Full details are confidential, but I am happy to walk through the problem, process, and system decisions in conversation.

Technodom Credit Flow Redesign: $4M+ monthly revenue impact->