# glvn.me Last updated: 2026-06-12 ## Site Summary glvn.me is a lightweight, Markdown-driven personal site and blog for glvnme. It presents product engineering, interface work, AI prompt workflow projects, skills, and short writing. The site is intentionally public. Public pages and Markdown source files may be crawled, indexed, summarized, and cited with attribution to https://glvn.me/. ## Canonical URLs - Home: https://glvn.me/ - Blog: https://glvn.me/blog/ - Contact: https://glvn.me/contact/ - Sitemap: https://glvn.me/sitemap.xml - Robots policy: https://glvn.me/robots.txt - LLM index: https://glvn.me/llms.txt - Legacy LLM pointer: https://glvn.me/llm.txt - Agent guide: https://glvn.me/agents.txt - AI context: https://glvn.me/ai.txt - Humans file: https://glvn.me/humans.txt - Content authoring spec: https://glvn.me/docs/content-authoring.md ## Home Metadata > X (https://x.com/glvnme) GitHub (https://github.com/glvnme) LinkedIn (https://linkedin.com/in/glvnme) Schedule a call (https://cal.com/glvnme) ## about Source: https://glvn.me/content/home/sections/about.md I'm Victor Golovin, an AI enthusiast and computational designer building software, tools, and automated workflows for architecture, engineering, and tensile fabric structures. My core expertise is parametric design and computational analysis for architectural and engineering processes. A concrete example is a full stadium roof and facade system: instead of manually redrawing every option, I define the design logic once - bowl geometry, roof edge, mast and grid rules, membrane bays, facade rhythm, support points, material constraints, analysis inputs, patterning, quantities, and drawing outputs - then let the connected models update together when the driving parameters change. That is the kind of work I mean by computation: moving design decisions into systems that can generate, check, compare, and document complex variants without losing control of the architecture. My background is tensile fabric structure design: membrane roofs, lightweight enclosures, detail-heavy coordination, and the hard translation of a strong concept into a buildable system. Now I am turning that background into digital services and software for the tensile fabric structures industry. The path is deliberately hands-on: build small tools, ship them, test features in development and production, learn through real workflows, and turn the useful parts into products. I work AI-native across different models, coding agents, automation harnesses, prompt systems, scraping flows, databases, image and media tools, and chat interfaces. I care less about demos and more about process improvements that save time, preserve knowledge, and make specialist work easier to reuse. The first focus is my own industry, because I know its constraints firsthand. Over time I plan to expand that work into broader software development: useful interfaces, internal tools, business-process automation, content systems, and practical AI-assisted products. More current experiments live in loading... (#loading), where I keep the things I am building before they are ready for a full writeup. background: - Stadium-scale parametric design systems for architecture, engineering, roofs, facades, and membrane structures. - Tensile fabric structure design, roof and facade systems, architectural membranes, and project coordination. - AI-assisted software creation, process automation, prompt systems, and tool testing. interests: - Software for membrane structure design, visualization, documentation, and delivery. - AI tools that support real work instead of producing disposable demos. - Databases, scraping, media generation, chat UX, agents, and automation ecosystems. next steps: Start with tools for tensile fabric structures, because the problems are specific and familiar. Use those projects to gain production experience, test what actually helps, and build a practical software portfolio. Then widen the scope into general software development and digital services for other fields. ## we work Source: https://glvn.me/content/home/sections/we-work.md Focused product, interface, and AI help for teams with a clear problem, a messy workflow, or business knowledge that should become easier to use. focus: - AI and LLM consulting for text, images, video, automation, agents, internal knowledge, and business-process improvement. - Product engineering, frontend interfaces, static sites, content systems, prototypes, and small web apps. - Current development interests: prompt systems, coding-agent skills, local-first tools, media workflows, business automations, and design/construction software. process: You bring a problem, process, or idea. We discuss the goal, constraints, and examples. I suggest a small first target, we test it in real use, then expand, narrow, or stop based on what helps. ## awen-prompt Source: https://glvn.me/content/home/sections/awen-prompt.md AwenPrompt is a local-first workspace for writing better architectural visualization prompts. I built it around the details generic prompt tools usually miss: reference-image geometry, tensile fabric behavior, real material colors, camera intent, and clear rules for what should not change. It does not generate images. It helps prepare copy-ready prompts for external image tools, especially for tensile structures, membrane, ETFE, mesh, facades, interiors, site scenes, and lighting studies. Open the production app (https://awen-prompt.vercel.app/) or read the longer blog article, AwenPrompt: structured prompts for architectural visualization (https://glvn.me/blog/#awenprompt-structured-prompts-architectural-visualization). why i made it: I kept seeing AI render workflows make the same architectural mistakes: masts disappeared, cables moved, fabric colors drifted, roof curves flattened, and reference images became mood boards. AwenPrompt is my way of making the important constraints explicit before the prompt leaves the design workflow. quick overview: - Prompt builder for tensile structures, facades, interiors, sites, and lighting studies. - Auto-Fill from a brief, reference image, or existing prompt. - Material catalog for fabric, structure, and architectural finishes. - Native mode that works without an API key. - Optional AI-assisted mode through hosted demo or bring-your-own-key routes. - Mask and lighting editors for targeted image edits. - Browser-local workspace for sessions, reviews, import, and export. prompt flow: Start by choosing the kind of architectural study you are making. AwenPrompt then shows the fields that matter for that workflow: site context, materials, fabric or envelope behavior, camera, lighting, output style, preservation rules, and negative prompts. You can fill the fields by hand, let Auto-Fill read a brief or reference image, or reuse an older prompt. The app then assembles the result as text or JSON so it can be copied into the image tool you actually use. fallback feature: Native mode is the main fallback. Once the app loads, it can assemble prompts from the form without an account, hosted demo quota, or API key. The hosted AI route also tries a short server-owned fallback chain for temporary provider failures. If that still fails, the user can keep working in Native mode or use a personal provider key for the current browser tab. tech stack: - Next.js, React, and TypeScript. - Tailwind CSS with shadcn/ui, Radix primitives, and lucide-react icons. - OpenRouter hosted demo routes, plus OpenRouter and OpenAI BYOK support. - IndexedDB and localStorage for browser-local workspace data. - Vitest and ESLint for checks around domain logic, routes, providers, storage, and UI behavior. ## golovin.design Source: https://glvn.me/content/home/sections/golovin-design.md golovin.design (https://www.golovin.design/) is my portfolio for computational design and tensile fabric structures. I have spent 12+ years working on complex tensile fabric structures, with 10 international collaborative project awards, 500+ completed projects, work across 9 countries, and a practice focused on turning ambitious membrane architecture into buildable spaces. projects: The project gallery collects selected tensile fabric, membrane architecture, ETFE, PTFE, PVC, and mesh structure work. It includes public squares, amphitheaters, stadia, hospitality spaces, retail canopies, playgrounds, and experimental studies, with each entry showing project context, location, year, collaborators, images, and award notes where available. services: The services section explains the full design cycle I provide: concept and schematic design, computational form-finding, tensile fabric analysis, structural detailing, patterning, fabrication documentation, construction drawings, material specification, tender support, installation coordination, and custom AEC workflow tools. overview: The site also documents computational design support, AEC software development, experiments, publications, and field notes. The core offer is a blend of architectural design judgment, parametric modeling, simulation, fabrication awareness, and practical coordination for teams that need membrane structures and design tools to turn early intent into production-ready work. ## skills Source: https://glvn.me/content/home/sections/skills.md glvn-skills (https://github.com/glvnme/glvn-skills) is my public library of AI skills: short, reusable instructions that teach coding assistants how I want them to handle specific kinds of work. For a non-technical reader, a skill is like a checklist, briefing note, and quality standard in one file. It exists because AI tools are more useful when they inherit a careful working method instead of starting every task from scratch. frontend design: This skill helps an assistant build cleaner sites, apps, dashboards, and prototypes. The benefit is not just prettier screens: it gives the work a stronger starting point with clear hierarchy, useful controls, readable layouts, and fewer generic AI-looking patterns. ai dev hardening: This skill turns a quick AI implementation into something closer to ship-ready work. It asks the assistant to check evidence, tests, edge cases, error handling, and unfinished shortcuts before calling the job done. maintain project context: This skill keeps the instructions around a project current and small enough to be useful. That matters because stale context makes AI assistants confidently repeat old assumptions; maintained context helps them understand the real product, current code, and decisions already made. ## loading... Source: https://glvn.me/content/home/sections/in-development.md Short notes on projects and experiments that are still in motion. membranium: Membranium is the Next.js and Convex scaffold for a tensile fabric registry: one web app with route-owned modules for Atlas, projects, magazine, companies, products, resources, supply, accounts, and admin. The local docs frame it as a modular monolith instead of separate deployments, with a small app shell, module-owned read models, and Convex as the data center for registry records, search, AI jobs, saved items, and editorial work. membranes.wiki: Membranes.wiki is the private admin and AI wiki factory for a public encyclopedia of tensile fabric structures. It keeps source discovery, drafts, revisions, reviews, taxonomy, source metadata, SQLite factory state, and export controls in a private Next app, then only published, public-safe pages can move into the public site through explicit export. otto: Otto is a Chrome-first WebGPU 3D editor for local-first sketch, procedural, CAD, and AI scene-patch workflows. It uses Three.js WebGPU, command-based undo and redo, deterministic project JSON, local .cym3d bundles, optional Convex persistence, worker-backed OpenCascade jobs, and a strict split between authored state and runtime previews. cymatic / eigenmode: Cymatic / Eigenmode is the tensile-practice operating dashboard. Cymatic holds the wider product map for leads, jobs, projects, tasks, time tracking, library, media, templates, contractors, financials, analytics, and AI settings; Eigenmode is the trimmed active dashboard built on Bun, Next.js, React, TypeScript, Tailwind, Biome, Convex, and an optional Electron wrapper for local desktop use. ImaGenX: ImaGenX is a node-based canvas for AI image and video pipelines. Users connect typed nodes for image input, annotation, prompts, multi-part prompt assembly, model generation, text generation, and output review, with provider support for Gemini, OpenAI, Replicate, and fal.ai plus local workflow JSON import and export for repeatable visual experiments. rhino-connect: Rhino Connect is a local-only connector that lets Codex talk to an active Rhino 7 session without screen automation. A RhinoCommon plugin exposes a token-protected API on 127.0.0.1, and a small MCP server offers tools for document state, selection, object listing, structured geometry ops, allowlisted Rhino commands, imports, and viewport captures. ndn-workflows: NDN Workflows is the research and architecture track for reading Membrane 9.10 VS 2017 .ndn files and bringing them into Rhino. The docs define a lossless AST plus normalized JSON projection, validation rules for nodes, elements, forces, layers, pattern options, seam data, and a future Rhino 8 plugin shape for import, metadata, refresh, .out analysis review, result coloring, and selection sync. ## misc Source: https://glvn.me/content/home/sections/misc.md Smaller projects and one-offs that don't fit elsewhere. ## Blog: If AI Can Render It, Build What It Can't Source: https://glvn.me/content/blog/posts/2026-06-12-if-ai-can-render-it-build-what-it-cant.md Date: 2026-06-12 Topic: AI Summary: AI is flooding the digital layer around products and services, making delivery, trust, operations, and real product experience more important. If AI can reliably generate visuals of your work, the move is not to panic. The move is to adjust. A lot of businesses are still treating this as a content problem. It is bigger than that. When images, videos, renders, prototypes, and eventually 3D product experiences become cheap to produce, the way customers judge businesses starts to change. The pipeline now runs through text-to-image, video, world-model experiments, and usable 3D models with fewer steps each month. That does not mean every physical product or real service disappears. It means the digital layer around those products and services gets flooded. Everyone will be able to look more professional than they actually are. Mediocre competitors will have beautiful visuals, polished ads, clean mockups, automated messages, and websites that make them look bigger, faster, and more capable than they really are. The Visual Moat is Evaporating: If your competitive edge is that you can produce a slick render, a polished mockup, or a gorgeous concept board, I need you to hear this: that is no longer an edge. It is table stakes. Soon, it will be cheap enough to stop mattering by itself. That creates a real problem for serious operators. Low-quality businesses can use better digital presentation to enter markets, undercut pricing, and overpromise. They may underdeliver, but by the time the customer realizes it, the market has already absorbed the price pressure. Margins get compressed. Trust goes down. Clients become more skeptical. Good companies are forced to spend more energy explaining why quality costs more. I have seen this happen in Mexico many times. A market gets flooded with low-quality offerings that look good enough externally. They sell aggressively, deliver poorly, and still manage to drag everyone's pricing down. AI makes that pattern easier to scale because the surface area of the business can now be manufactured faster than the actual capability behind it. If your only differentiator is the digital wrapper, you are about to be undercut by automation. Get to the Production Floor: So the answer is not simply "use AI." The answer is to strengthen the parts of the business that are harder to fake. You run to the factory floor. The supplier list. The bill of materials. The shipping workflow. Fix the process that actually delivers. Can you find a cheaper supplier without sacrificing quality? Can you find a better material that costs the same? Can you reduce handoffs, speed up quoting, remove failure points, or make installation less painful? Have you actually pressure-tested your product-market fit lately, or are you assuming the market still wants what you built three years ago? Rethink your business structure around what clients actually pay for. Not what your deck says matters. What they return for and tell their friends about. Sometimes the thing they buy is not only the product. It may be reliability, speed, installation, maintenance, customization, accountability, or the fact that the client can trust you to actually deliver. Make Quality Visible: At the same time, the digital side cannot be ignored. Distribution matters. Online presence matters. Clear positioning matters. Your quoting process, onboarding, follow-up, documentation, and back-office systems matter. A competitor with worse execution but a better digital experience can still take attention, leads, and pricing power away. Being good is not enough if the customer cannot see it before the sale. The real opportunity is to combine both sides: a strong analog product experience with a strong service and digital experience around it. Physical quality becomes more valuable when the market is full of artificial polish. But quality has to be visible, understandable, and easy to buy. Let AI handle the plumbing where it helps. Automate the spreadsheet. Generate the draft ad copy. Render the concept art for internal review. Improve customer support. Test ideas cheaper. Reduce admin work. Use it. But do not confuse the lever with the load. The New Premium is Analog: Here is the counterintuitive win: as generative visuals flood the market, real becomes rare. And rare becomes valuable. The analog product experience, physically holding something well-made, interacting with a human who knows the details, receiving service that wraps the product in context and care, that is where pricing power lives. The world-model can generate the idea of a chair. It cannot account for the grain of the wood under your client's palm six months after delivery. Your competitive moat is not the render. It is the reality behind the render. The Canopy: This is not really about AI. You do not need to turn your company into an AI company, and you do not need to adopt every new tool. The bigger point is that cheaper digital presentation changes how services are discovered, compared, sold, and trusted. When that happens, the operating logic of the business has to change with it. Imagine the news says a hurricane is forming and will hit land in two weeks. The correct response is not fear. It is preparation. Reinforce what matters. Remove what is weak. Protect the parts of the business that customers can verify after the sale. Or imagine a new forest, full of young trees competing for sunlight. Some are larger, some are smaller, but all of them are reaching for the same space. The ones that survive will not be the ones with the prettiest leaves. They will be the ones with the deepest roots and the most efficient systems for pulling water out of the ground. The surface is getting easier to fake. Build the part that still has to be real. ## Blog: AwenPrompt: structured prompts for architectural visualization Source: https://glvn.me/content/blog/posts/2026-06-11-awenprompt-structured-prompts-architectural-visualization.md Date: 2026-06-11 Topic: AwenPrompt Summary: A practical introduction to AwenPrompt, a local-first prompt builder for architectural visualization workflows that need reference-image and material fidelity. AwenPrompt is a local-first browser tool for architects and designers who need better prompts for architectural visualization. It is strongest when the project involves tensile fabric, membrane structures, ETFE, mesh, shade systems, masts, cables, and lightweight envelopes. It also supports facade studies, interior visualization, site scenes, and lighting studies. Open AwenPrompt (https://awen-prompt.vercel.app/) to try the production app. Embedded video: https://player.vimeo.com/video/1200076771?context=Vimeo%5CController%5CApi%5CResources%5CVideoController.&h=dcdc11f9db&s=b023c42016f2574dfaa7c048346389a4505ab430_1781310715 The important part is what AwenPrompt does not do: it does not generate images. Instead, it helps turn design intent into structured prompt instructions that can be used in external image-generation tools. Its job is to keep the project legible when an AI image model starts interpreting it. Why it exists: Architectural prompts fail in predictable ways. A model may redraw the structure, invent supports, change fabric colors, flatten a roof form, ignore cable logic, or turn a careful reference image into a loose mood board. That is irritating for any visualization workflow. For tensile fabric and membrane architecture, it can break the whole idea, because the geometry, supports, curvature, and material behavior are not decorative details. They are the design. AwenPrompt treats those details as first-class inputs. Instead of asking users to write one long prompt from memory, it gives them a structured prompt builder. The user can describe the architectural workflow, site context, fabric system, material choices, camera view, lighting intent, output style, preservation rules, and negative prompt guardrails in a repeatable way. The result is not a magic sentence. It is a clearer brief for the next tool in the chain. What it does: AwenPrompt is built around a few practical functions: - Prompt builder: a structured input surface for architectural visualization prompts. - Architectural workflows: presets for tensile structures, facade studies, interiors, site work, and lighting studies. - Native mode: deterministic prompt assembly from the form state, without calling an AI provider. - AI-assisted mode: provider-backed prompt generation through server-side routes when a configured OpenRouter or OpenAI key is available. - Reference image analysis: image-based project reading for geometry, composition, material, and context clues. - Auto-fill: conversion of a brief, reference image, or existing prompt into prompt builder fields. - Strict overrides: authoritative user fields that AI-assisted output must preserve. - Negative prompts: guardrails against geometry distortion, wrong fabric colors, unrealistic physics, and low-quality rendering artifacts. - Material catalog: built-in fabric and architectural material data, with support for custom materials. - Mask editor: region-based inpainting instructions for targeted changes. - Lighting editor: lighting annotations that can be exported as prompt text. - Local workspace: browser-local sessions and artifacts, with no account required. How the workflow feels: A typical session starts with a design problem rather than a blank text box. You choose the architectural workflow, then describe the project: structure, site, view, materials, atmosphere, output style, and anything that must not change. If there is a reference image, you can use it to anchor massing, camera angle, fabric topology, openings, site boundaries, or existing context. From there, AwenPrompt can produce either a deterministic Native mode prompt or an AI-assisted prompt. The output can be copied as text or structured JSON, depending on the downstream image model target. If the image needs a local change, the mask editor helps define bounded regions and inpainting instructions. If the task is about mood, fixture placement, or emphasis, the lighting editor lets you annotate lighting areas and export a focused lighting prompt. The app is not trying to replace design judgment. It gives the user a tighter way to communicate that judgment to image tools. What makes it different: Most prompt tools are broad. They help with style, mood, subject matter, and composition, but they do not know much about the failure modes of architectural visualization. AwenPrompt is narrower on purpose. It cares about whether a membrane roof keeps its topology. It cares whether the fabric color stays the selected manufacturer color instead of drifting into a nearby shade. It cares whether a reference facade keeps its rhythm, openings, grid, and camera view. It cares whether a negative prompt is specific enough to stop the common damage. That narrowness makes it more useful for the work it is built for. Built for local-first work: AwenPrompt is designed to stay usable without an account. Project data is kept in the browser. Native mode does not require an API key. AI-assisted routes use server-side provider handling, and user-facing workflows are written with the assumption that model output is untrusted until parsed and validated. That matters for design work. Reference images, prompt drafts, and project notes can be sensitive even when they are not formally confidential. The app keeps the normal workflow lightweight while avoiding unnecessary server-side project storage. Who it is for: AwenPrompt is for people who already know what they are trying to see. It is useful for architects testing facade options, designers preparing concept visuals, tensile fabric specialists trying to preserve form and material intent, and anyone using reference images as serious constraints rather than loose inspiration. It is also useful for teams that need prompts to be repeatable. A structured prompt builder makes it easier to see why one output worked, why another failed, and which field needs to change before the next run. Current status: AwenPrompt is in active development. The core product surfaces are the prompt builder, Auto-fill workflows, reference image analysis, mask editor, lighting editor, and local workspace. The current focus is not to make the broadest possible image tool. It is to make a careful prompt workspace for architectural visualization, especially where geometry, fabric behavior, material color, and reference-image preservation are the difference between a useful study and a pretty but unusable picture. ## Blog: Test post: opening the notebook Source: https://glvn.me/content/blog/posts/2026-06-08-test-post.md Date: 2026-06-08 Topic: Site notes Summary: This is a placeholder entry for the new blog route: short, structured, and ready to be replaced by a real note. The blog page is meant to keep writing close to the rest of the site: direct text, durable links, and a layout that makes older posts easy to scan by date and topic. A single registry on the left can grow into a compact archive without changing the reading surface. Each entry should carry a date, a topic, and a plain title so the archive stays useful when there are more posts. What belongs here: - Implementation notes that are useful after the work ships. - Small interface studies and product decisions. - Project updates that deserve more context than a link. For now, this filler post proves the page shape: registry on the left, writing on the right, and a route that can be expanded without adding a build step. ## Access Policy This is public promotional and editorial content. Search engines, AI retrieval crawlers, and user-triggered agents may access public pages and public Markdown sources. Prefer citing the canonical home URL or the specific Markdown source URL used as evidence. ## Content Authoring Contract The canonical authoring instructions live at https://glvn.me/docs/content-authoring.md. Home sections are managed through `content/home/index.json`, `content/home/meta.md`, and one Markdown file per section in `content/home/sections/`. Home source URLs: - https://glvn.me/content/home/index.json - https://glvn.me/content/home/meta.md - https://glvn.me/content/home/sections/about.md - https://glvn.me/content/home/sections/we-work.md - https://glvn.me/content/home/sections/awen-prompt.md - https://glvn.me/content/home/sections/golovin-design.md - https://glvn.me/content/home/sections/skills.md - https://glvn.me/content/home/sections/in-development.md - https://glvn.me/content/home/sections/misc.md Blog posts are managed by dropping Markdown files into `content/blog/posts/`. The browser uses generated same-page archive data in production, with generated Markdown alternate links as a static fallback. Blog source URLs: - https://glvn.me/content/blog/posts/2026-06-12-if-ai-can-render-it-build-what-it-cant.md - https://glvn.me/content/blog/posts/2026-06-11-awenprompt-structured-prompts-architectural-visualization.md - https://glvn.me/content/blog/posts/2026-06-08-test-post.md The renderer supports a small Markdown subset: paragraphs, `- ` lists, inline links, inline code, bold text, and blog body `##`/`###` headings. Markdown links only become anchors for `http:`, `https:`, and `mailto:` URLs.