# Viskhan Khasiyev > Design engineer in Antwerp and a professional maker across design, code and automation. Currently at Bundl, with recent client involvements including Nestlé, Lilly, Chanel and Gore. This file contains the full content of https://khasiyev.com as one markdown document. The link index lives at https://khasiyev.com/llms.txt. --- # OBSN Section: Client work Published: 2026-09-14 Role: Creative Developer Clients: Steven.com, Authentic Brands Group Live site: https://obsession.com Case study: https://khasiyev.com/work/obsn Obsession is a joint venture between Steven.com, Steven Bartlett's creator-economy holding company, and Authentic Brands Group: media, events, a network and a venture arm under one mark, OBSN. The site is the launch platform, live from 14 September 2026 with one call to action, join the list. My part was the top of the homepage: a five-frame scroll story the team calls the rewind. The video above is the fold with the interface hidden, the world on its own. ## The rewind Every creator started with one thing. The fold takes that line literally and runs a creator's career backwards. A follower counter starts at 1,000,000 and counts down to zero, one room per scroll stop. The eclipse first: the O mark as a sun on the bottom edge of the screen, alive under the pointer. Scrolling dives into the disc. Before the arena, a wireframe stadium with a drone, cameras and a mic hurtling past the lens. Before the team, a studio: ring light, two chairs, headphones, a notebook, an iPad. Before anybody was watching, the bedroom: MacBook, LED lamp, USB stick, post-its. The scroll stops once more at the bedroom door as it cracks open, then the last swipe lands on the line the page was written for. There was one thing: obsession. The colour runs as a single vortex from scene to scene, amber in the arena, Absolute Orange in the studio, deep red in the bedroom, ember at the landing. ## The build Built with agentic coding on top of the team's Next.js 15 repo: React 19, three.js through React Three Fiber, drei and postprocessing. The preloader counts 000 to 100 on what the first frame actually needs while the O mark assembles, so the sun is formed by the time the page hands over. ## The same story on a phone Effect tiers per GPU, full, mid and lite, so a phone plays the same five rooms with a lighter pass. A Reduce motion switch sits in the header for anyone who wants the story without the flight. --- # OBSN at Lightbox NYC Section: Client work Published: 2026-09-13 Role: Creative Developer At: Lightbox NYC, Manhattan Clients: Steven.com, Authentic Brands Group Case study: https://khasiyev.com/work/obsn-lightbox Obsession launched on 14 September 2026 with a night at Lightbox NYC, a room in Midtown Manhattan whose four walls are one 360° projection surface. The venue piece came to me with a day to go: a loop of about a minute for the four walls, delivered as four files cut to the venue's wall templates, due the next afternoon. The video above puts the finished walls back into a model of the room. ## One world at a time The loop is the fold from the site, taken off the screen and wrapped round the room. You stand inside the arena, the crowd twinkling on every wall, the drone on its lap overhead. Then the room dissolves into a field of particles, the eclipse closes onto the primary wall with the next world already forming inside its disc, opens again, and the studio draws itself on around you, line by line. Twenty seconds later the same pass takes you to the bedroom, and once more back to the arena, so it never ends. Props glow one at a time while the rest stays dim, which does the work of a spotlight without one. Nothing hurtles and nothing is written on the walls. It plays behind people holding a drink, so every move is slow enough to follow from anywhere in the room. ## Four walls from one camera The four files are a box render: one camera at the centre of the room, four views at ninety degrees, the venue's own proportions setting the fields of view, so the picture bends at the corners exactly where the walls do. The eclipse is the site's own shader with its screen coordinate swapped for a projection of each pixel's direction in the room, which keeps the disc a true circle from the centre and lets the corona run round the corners. --- # 113 Spring Section: Client work Published: 2026-01-01 Role: Designer & Developer At: Bundl, Antwerp Live site: https://113spring.com Case study: https://khasiyev.com/work/113-spring 113 Spring is an undisclosed brand built around curated products, experiences and events. At Bundl I worked as designer and developer on the production Shopify storefront, together with [Montana Mertens](https://www.linkedin.com/in/montanamertens/), Bundl's senior designer, who led the design. My side was taking the Figma design system through to the custom theme code live at 113spring.com. ## The storefront The site carries an editorial calm that is easy to sketch and hard to keep in a real store: full-bleed film stills, a restrained palette, and typography doing most of the work. That calm is Montana's design; building the theme meant her quiet details survived contact with production, from the navigation rhythm down to the product grid. ## A theme that scales past launch Underneath, the build is the same scalable component system as OLEUS: pre-built, configurable sections the team composes into new pages in the Shopify editor. The calm holds because the system enforces it, page after page, campaign after campaign, with no developer needed for the next landing page. ## Details that carry the brand The preloader film above is one of those details. Small moments like it set the tone before a single product appears, and they only ship when design and build stay close. ## Credits Shout out to [Montana Mertens](https://www.linkedin.com/in/montanamertens/) for the great design work that carries this storefront, and to [Jan Van Fraeyenhoven](https://www.linkedin.com/in/jan-van-fraeyenhoven/) for setting up all the tracking behind it. --- # OLEUS Section: Client work Published: 2025-05-01 Role: Designer & Developer At: Bundl, Antwerp Clients: Nestlé Live site: https://oleus.com Case study: https://khasiyev.com/work/oleus OLEUS is a Nestlé venture: mitochondrial health shots rooted in Swiss science. Through Bundl I joined as designer and developer on the production storefront, working alongside [Montana Mertens](https://www.linkedin.com/in/montanamertens/), Bundl's senior designer, who led the design, from the Figma design system to the custom Shopify theme code running at oleus.com. ## From Figma system to theme code The storefront runs on a fully custom Shopify theme. Montana led the design; my work sat on both sides of the usual handoff, extending pages in her Figma system and building them as Liquid sections and templates. Staying close to both ends meant the design intent survived all the way to production, down to the type scale and the easing curves. ## A system the team can build with The theme is a scalable component build, not a set of hard-coded pages. Every section ships as a reusable, configurable unit with its own schema, so the brand team assembles new pages and campaigns straight in the Shopify editor. Layouts stay on brand without a developer in the loop, because the guardrails live in the components themselves. ## Built to sell, not to demo This is a live store with real customers, real product data, and a review count that keeps moving. Every design decision had to hold up against changing content and a running catalog, not a frozen staging environment. That constraint shaped the system more than any moodboard did. ## Credits Shout out to [Montana Mertens](https://www.linkedin.com/in/montanamertens/) for the great design work this storefront is built on, and to [Jan Van Fraeyenhoven](https://www.linkedin.com/in/jan-van-fraeyenhoven/) for setting up all the tracking behind it. --- # Venture Design at Bundl Section: Client work Published: 2025-03-01 Role: Venture Designer & Developer At: Bundl, Antwerp Clients: Gore-Tex, Lilly, OLEUS (Nestlé), Chanel, Liberty Global Case study: https://khasiyev.com/work/bundl-venture-design At Bundl, I work as a Venture Designer & Developer embedded in fast-moving venture sprints. Each engagement takes a strategic idea and turns it into something real: a brand, a prototype, a go-to-market kit, or all three at once. ## Client Involvements
Gore-Tex Lilly Oleus Nestlé Chanel Liberty Global
## What I do The work spans design, development, and strategy simultaneously. At Bundl there's no clean separation between the three: you're shaping the strategy while designing the site while building the product. That's both the difficulty and the part I find most interesting. **Design.** Web design, landing pages, and a high volume of ad visuals. AI video and image generation, video editing, email marketing. The goal is always the same: make the strategy visible and credible. **Development.** This has grown from MVPs and prototypes into production builds, fully custom Shopify themes included. More recently we build everything from scratch: Next.js for the web, Swift, React Native and Electron for native apps. With AI in the loop, the bottleneck on how much of that I can carry alone has vanished. **Strategy.** Venture sprints are too fast for a strategy-to-design handoff. I'm in the strategy conversations from the start, so the design and the build never have to guess why the venture exists or who it's for. ## AI workflows This is a newer part of the work, but one that's growing fast. I've been integrating AI into both the design and development side of sprints: **Agentic coding** sits at the core of how I build: CLI-first workflows, **MCP servers** that connect agents to the tools and data a project lives in, and **knowledge management systems** that keep context persistent from sprint to sprint. This space moves faster than any sprint cycle. People sometimes ask how I stay up to date with it, and the answer is that I don't have to try: pushing AI hard through every piece of work keeps me current automatically. The rest of the job is knowing when to use it. ## What it actually looks like Each sprint is different. Sometimes I'm the only designer and developer on the project, covering everything from the first design exploration to the deployed landing page. Sometimes I'm working alongside strategists and project leads, translating their work into something visual. What doesn't change: the expectation that every problem is solvable, the pace is fast, and the output has to be good enough to put in front of serious stakeholders. That pressure is something I've come to appreciate. --- # Tesla Blender Project, remastered Section: Lab Published: 2026-09-27 Case study: https://khasiyev.com/lab/cybertruck-remaster In 2024 I spent three weeks on my first ever Blender project: a Cybertruck rolling out of a bunker on Mars. Modelling, texturing, animating, all by hand and all new to me. [The original is still here](https://khasiyev.com/university/tesla-blender). I was proud of it back then. Looking at it now, it's kind of wack. So I gave it to Opus 5.5 with one message: "couple of years ago i made this blender render and its kinda wack now. keep the story, the bunker, the domes, starship on the horizon, but rebuild everything else. the models, the scene, the animation!! it shouldnt feel like a 3d render, it should feel like a film someone actually shot on mars. heavy, quiet, a bit of dust in the air. this is not a short task, loop until you make something awards worthy". ## Everything is code It started by reading my old renders, the styleboard and the sketch sheet, and kept the story: nine shots, the bunker, the greenhouse domes, the solar field, the door, the drive-out, Starship on the horizon. Then it rebuilt all of it without once opening Blender's interface. The truck is cut from planes and checked against the side view in Tesla's owner's manual. The terrain is a heightfield, the rocks are scans scattered with geometry nodes, and the domes are geodesic glasshouses with lettuce growing in the racks. Even the soundtrack is synthesised in Python. That's the part that gets me. Nothing in this film was placed by hand. Around 60 scripts rebuild the whole scene from source, so any shot can be changed and rendered again. ## Where I steered I didn't write any code, but I didn't sit back either. My part turned into art direction. "the wheels look like plastic. rubber is matte and dusty, only the rims should catch the light." "the dust looks fake, real dust hangs low and trails behind the wheels, and the sun should come through it." "the animation is stiff. something this heavy should settle into its suspension when it stops, nothing moves at a perfectly even speed." Every note went back into the loop, and the next pass came back better. On the first day up to six sessions ran side by side: one on the truck, one on the wheels, one on the capsules, one on the hangar interior, one on the bunker gate and one stitching it all together. ## Post-processing For the finish I switched to goal mode, where you set a goal and the agent keeps looping until it's met. Those passes ran on Fable 5.1, with goals like "make it look shot, not rendered. think anamorphic sci-fi like dune and interstellar: deep blacks, the sun blooming a little too much, grain you feel more than you see". Instead of Blender's compositor it wrote its own finishing chain in Python, in the order a camera, a lab and a colourist would apply it: per-shot exposure and white balance, a graduated ND on the sky, lens glare, anamorphic streaks, red halation around the light bar, a log grade with a split tone, then grain. Drag to compare the raw render with the delivered frame. ![The same hangar frame at four stages of the finishing chain](https://khasiyev.com/images/projects/cybertruck-remaster/grade-breakdown.webp) The same frame through the chain: raw render, then exposure and white balance, then lens and halation, then the print grade and grain. ## Behind the scenes The whole film was made headless: no window, just scripts driving Blender from the terminal. Open the files in Blender afterwards and the scene is all there, with real node trees, real materials and all nine shot cameras on the timeline. The dust was the hardest part. The first version was 36 overlapping puffs that looked fake and took 40 of the 60 seconds each frame needed to render. It got replaced by a single volume, which looks better and renders faster. - About 46 hours from my first message to the final film - Around 2,100 tool calls, 27 messages from me - 60 Python scripts, close to 19,000 lines - 815 frames at 2.39:1, rendered on an M4 Max in about a day, with the final crane shot alone taking eight hours - Score and sound design synthesised from scratch, no samples - Scanned textures, rocks and skies from Poly Haven (CC0) Three weeks in 2024, two days in 2026. I wrote down [what that does to my view on AI](https://khasiyev.com/blog/where-i-stand-on-ai). --- # Portfolio v1 Section: Lab Published: 2025-01-15 Case study: https://khasiyev.com/lab/portfolio-v1 Before this site existed, I built a first take on my portfolio in Webflow. One dark, cinematic opening scene: a stretched wordmark sinking into the terrain, a ship drifting overhead, scanlines across the whole frame. ## Never shipped, still worth sharing It never went live. By the time the header was where I wanted it, my idea of what a portfolio should say had moved on, and this stayed a concept. I'm sharing it anyway, because it was a genuinely fun build and it still holds up. What stuck with me is what it did to my view of no-code platforms. This was my first real project in Webflow, and the entire scene came together inside the designer, without writing a line of code. That kind of speed changes how you prototype: you can chase an atmosphere instead of an implementation. It opened my eyes to how much these tools can carry on their own. --- # Tesla in Blender Section: University work Published: 2024-11-01 Case study: https://khasiyev.com/university/tesla-blender Third year. This is what I ended up with after three weeks in Blender, coming in with zero 3D experience. I picked Tesla as the subject because the geometry felt interesting: clean surfaces, deliberate proportions, good lighting potential. The project was whatever you made of it; I wanted to see how far I could push the render in three weeks. What followed was a crash course in everything simultaneously: modeling, materials, lighting setup, camera angles, render settings, compositing. All new, all at once. Devine had trained me to figure things out without being told exactly how, which turns out to be genuinely useful when the software is as deep as Blender. You can't learn it linearly, you start making things, hit a wall, figure out what you're missing, and keep going. That's mostly how this went. One thing that helped more than I expected: I had done some physical sculpting when I was younger. There's something about thinking in three dimensions that was already familiar even if the software wasn't. Getting geometry to sit right, understanding how light wraps around a curved surface, some of that translated in a way that surprised me. The result isn't perfect. But it's a real result, made in three weeks, with a tool I'd never touched before. I'll take it. ## What I learned 3D design requires a different mental model than flat design. You're not making something look right in one view, you're building something that exists in space and then choosing how to observe it. The lighting, the camera, the materials, the geometry all interact in ways you have to discover through experimentation. It's a good discipline to have and one I'd like to go deeper on. In 2026 I handed this project to an AI agent and let it rebuild the whole thing from code. [See the remaster](https://khasiyev.com/lab/cybertruck-remaster). --- # Scapino Ballet Rotterdam Section: University work Published: 2024-06-19 Role: Development Clients: Scapino Ballet Rotterdam Case study: https://khasiyev.com/university/scapino Scapino Ballet Rotterdam is a historic dance company with an audience that skews older and affluent. The brief asked for a product that uses the educational and social side of dance to reach people in their twenties, with one condition attached: research the problem first, then design and code a working solution. I worked on the development side of the answer, an interactive poster that runs on a webcam and a vertical screen. ## The research Desk research into Scapino's own channels and an online survey among twenty-somethings pointed the same way. Visual appeal and exposure were not the problem. The gap was personal relevance: nobody in that age group saw themselves in modern dance. Two personas and one how-might-we question came out of the analysis: how might we make people in their twenties connect with modern dance, so that they attend a performance at Scapino Ballet? Three motivators carried the concept from there: personal relevance, shared experiences and authentic storytelling. ## The concept A poster that talks first. It stands where the audience already is and opens with a dancer's story instead of a season programme. Would you quit a top-tier company to be free? She did. Put your hands up and she tells you how. Once the story lands, the poster turns the question around: how far would you go to chase your passion? The answer is not typed. It is drawn. ## Hands as the interface The poster runs as a web app inside an Electron kiosk shell, with ml5.js reading the visitor's hands through a webcam above the screen. There is no touch and no menu. A thumbs up or down answers a yes-or-no, holding both hands together for three seconds starts the generator, and the distance between two open hands drives the artwork: the further apart, the longer the word FAR stretches and the more brush strokes land on the canvas. Every gesture shows a visible timer so people know the poster is listening, and a marker on the floor tells them where to stand. ## Art you can take home The session ends with the visitor's own artwork on screen and a ticket coming out of a printer beside the stand. The ticket carries the artwork, the sentence they made and a QR code to a companion website, where the dancers' stories continue and a performance ticket is one tap away. A physical object made in public is the kind of thing people show to friends, which is exactly the shared experience the research asked for. ## Tested in the room We tested it ourselves first, then with users on the real setup, and changed the build after each round. The clip below is the final prototype from testing day. --- # Mixbox Section: University work Published: 2024-06-01 Case study: https://khasiyev.com/university/mixbox End of year two, and this was the final design project. Mixbox was about stepping back from single-deliverable projects and thinking in systems. The main focus was app design, but you couldn't design the app in isolation, you had to build the brand around it first. Visual identity, packaging, typography, color, the whole system. Only once that was defined did the app screens actually make sense as part of something coherent. The brand constraints you create early become the guardrails that make the design decisions easier later. That shift in scope is what made this project different from everything before it. It wasn't "make screens that look good", it was "figure out what this thing is, and then make it." That's a harder problem and a more honest one. ## What I learned Brand systems and product design aren't separate disciplines. The better you define the former, the more consistent and intentional the latter becomes. This was the first project where that relationship was the whole point, and working through it from scratch made the logic of design systems click in a way it hadn't before. It also made clear how much energy gets wasted when you try to skip that step. --- # Larq Kickstarter Section: University work Published: 2024-03-15 Case study: https://khasiyev.com/university/larq-kickstarter Mid second year, deeper into After Effects. Much deeper. We'd done the pinball machine the semester before. That was an introduction. This was the real thing. The assignment was to produce a full Kickstarter-style motion graphics video for a product of your choosing. I picked Larq, a self-cleaning water bottle. The process was the full production pipeline: choose a product, develop the concept, sketch it rough, create still frames, build the motion. Nothing gets skipped. The still frame stage is more important than it looks: you're essentially storyboarding in high fidelity before anything moves, which forces you to commit to timing and transitions on paper before you invest the time animating them. Getting used to that workflow took time. Motion graphics at this level isn't just making things animate, it's thinking in sequences, building a visual language first in stills and then bringing it to life in a way that feels deliberate rather than busy. I've always liked the Vox YouTube approach to motion: clean movement, data-driven visuals, editorial quality. Some of that found its way into the Larq video. Not a copy but you can see the influence.