Vazgen Harutyunyan

Vazgen Harutyunyan

Design Engineer

I bring product, design, and code together until they feel like one piece.

Selected work

03

Lumenarc

Lumenarc builds AI demo agents that learn a product and run live sales demos around the clock. We're piloting now with B2B teams who want to automate and streamline their sales funnels.

I'd worked with the founder on an earlier project. When he decided to make this one serious, I joined full-time. My title is principal product designer, but the work is closer to a design engineer than a pixel pusher. I built the brand from scratch and own the product UI end to end.

The hardest part was showing the agent without giving it a face. The mark itself had to carry presence, motion, and voice. The arcs and diamond combine into something that feels alive without being a character, and that is the piece I'm proudest of.

Lumenarc
Lumenarc · 02

This is the qualification UI the agent runs before each demo. The layout came out of Figma, but to make it feel right in motion I needed to tweak it live. So I built the debugger on the right. It lets me override avatar animation states, user voice intensity, transition timing, easing, and every other parameter without leaving the running app.

This is a pattern I use constantly now. Instead of going back to Figma to adjust a value and re-export, I build a tool that exposes the variables I care about inside the page itself. I started doing this in Cursor when I first got serious about AI and code, and now I'm working almost entirely in Claude Code. More of my design work happens in code than in Figma these days, and this debugger is one of the clearest examples of why.

Lumenarc
Lumenarc · 03

Shot shows the live demo experience from the user's side. Lumen runs a pre-recorded version of the client's product and walks the user through it the way a salesperson would. You can see the Lumen pointer hovering over the Scheduled tab, guiding where to look next. The agenda card on the lower right shows where the user is in the demo flow and what is coming up.

Lumenarc — 3 - Lumenarc
Lumenarc
Lumenarc · 04

This shot shows the agent moving through its different states. I built the animations from scratch for listening, thinking, and idle, then combined them so the mark itself signals what the agent is doing. The same system carries across every Lumenarc deployment, which means clients can name their agent something other than Lumen and the brand still holds. That call gives Lumenarc more brand exposure than a single humanoid avatar ever would.

idle
Lumenarc
Lumenarc · 05

This is the studio side of Lumenarc, where the client sees what happened on every session the agent ran. Each session opens up to show an AI-generated highlights summary, every objection the agent handled, the call recording, sentiment scores, talk ratios, and a recommended next step.

The point of this view is the feedback loop. Everything the agent did gets captured and broken down so the human team can review it, and so the agent itself can improve from previous conversations. Objections that came up, moments that landed, prospects worth following up on, all of it is structured for both human and agent to learn from. I designed the layout to be scannable in seconds because anyone running a demo program will be in this dashboard daily.

Lumenarc — 5 - Lumenarc
Lumenarc

ProSky

ProSky is a group flight management company. I've been working with them for almost three years and have shipped several features and side applications during that time.

This shot is from the latest one, cost reconciliation. I built it end to end, and it combines the accounting user with the project manager user that already existed in the platform. We are mid-rework on the whole product right now, and these shots are part of that effort.

The hardest part of cost reconciliation is the data itself. Billing comes from several sources and nothing can get lost in matching it back to the right project, booking, or passenger. I started by building a separate shell application just to figure out the reconciliation logic against real input files. During the handoff we agreed we did not need a shell anymore, and the developers were happy for me to work straight into production. Looking back, I think that was the moment that flipped the whole company toward AI-driven development, which is how most of the work happens here now.

ProSky — 1 - ProSky
ProSky
ProSky · 02

This is where the actual assignment happens. The accounting line from the airline's billing file sits on the left, and the user browses every booking and passenger inside the matched project on the right to find the one it belongs to.

The data is dense and the stakes are high. A single project can have dozens of passengers and multiple bookings, with several charges that look almost identical until you check the reference numbers. I designed this view so the user can move through matching work without losing their place. Source data stays pinned on the left, the matched project is summarized below it, and the browse list on the right is filterable by all, booking, or passenger. The assign button is the only commit, and once it fires, the whole line moves out of the unmatched queue.

ProSky — 2 - ProSky
ProSky
ProSky · 03

This is the project overview, before on the left and after on the right. Navigation has been an issue here for a long time. The data has only gotten denser as the platform has grown, and the old layout could not keep up.

The rework is not just a visual refresh. I'm building a clear design system underneath it so the next iteration of this view, and every other view, sits on the same foundation. No more one-off components, no more drift between pages that should feel like the same product.

After
Before
ProSky
ProSky · 04

Booking details is the densest screen in the product. The challenge is the same as everywhere else in the rework, just sharper. Show a huge amount of data without making the page collapse into noise.

The old version drifted over time. Several developers shipped to it over the years, and the design system that was defined back then did not hold. That is the failure mode I am designing against in this rework.

After
Before
ProSky

Stackdrop

Stackdrop is a development studio that builds internal tools on Retool and embeds forward-deployed engineers into client teams. They needed a marketing site that signaled engineering credibility, not generic SaaS polish.

I designed and built the marketing site end to end, covering brand direction, the illustration system, and the front-end.

The challenge was making "internal tools agency" feel precise rather than dull. I built the whole site as a technical blueprint. Isometric schematics, construction grids, and crop marks run through every section, so the page itself reads like the kind of meticulous engineering artifact Stackdrop produces for clients.

Stackdrop — 1 - Stackdrop
Stackdrop
Stackdrop · 02

This is the illustration system shown as a set. I built it so every concept on the site gets drawn the same way: isometric, on the construction grid, in the same line weight and palette, whether it's a connection diagram, a document grid, a database cluster, or a data signal.

Stackdrop — 2 - Stackdrop
Stackdrop
Stackdrop · 03

This is a single illustration from the system, shown at full size. The layered card construction stays in the isometric perspective the rest of the site uses, drawn in the same line weight and palette, sitting on the same construction grid.

Stackdrop · 04

This is from one of the UX consultations I ran while working with Stackdrop. drdm.io was one of their clients, a contract-management platform that had been built for its own developers, not for the outside companies about to onboard onto it. The original had redundant pages and a navigation structure that didn't match how new users would think.

Over a series of sync calls, I reorganised the full page architecture into clear, reusable groupings and defined the happy-path flows the team could build against going forward. The goal was to move the product from developer-ready to user-ready, aligned with the patterns its future users already expect.

I did similar consultation work for several other Stackdrop clients during that engagement. This one is included as a representative example of the UX reasoning and pattern thinking applied across that work.

After
Before
Stackdrop