Two products. Four surfaces. 800% token surge at launch.

Mapped the positioning of competing AI infrastructure and DePIN node sale projects. Three patterns emerged: pure infrastructure plays with no visible product; AI product plays with no visible infrastructure; node sale campaigns with no connection to the product. OSMI had all three components. The opportunity was to connect them coherently.

The marketing site had to serve two audiences without alienating either: crypto investors evaluating the node sale (care about tokenomics, network growth, reward mechanics) and AI users evaluating the product (care about model quality, features, and whether it beats alternatives). I designed the sitemap so each had a clear path through the site without being interrupted by content meant for the other.

All wireframes built with real copy. In a technical product with novel features, copy is architecture. A wireframe that says "headline here" cannot answer whether the page explains Truth Search to a first-time visitor. The wireframe pass resolved every content question before design began. Every page had a tested narrative before any UI work started.

POW Studio created the brand identity. I translated it into a site that felt like OSMI: technically serious, visually distinct. Dark backgrounds with precise technical typography, data visualization language in hero sections, interaction patterns borrowed from developer tools rather than marketing sites. The site was designed to signal: built by people who understand infrastructure. Full desktop and responsive UI across all marketing pages.

Four distinct capability layers designed as one coherent product. Core chat: the default conversation interface, clean and fast. Image Generation: visual creation integrated into the chat flow, not a separate tool. Truth Search: deep research mode with visible progress states and source cards, designed so a 30-60 second process feels intentional, not broken. Think: sequential reasoning mode with collapsible thinking steps and prominent final answer. Multiple iterations per feature before reaching production.

There was no tier selection in the checkout. The active sale tier was set globally by marketing as a campaign trigger. The checkout did one thing: show which tier was currently active and how many licenses were left. That simplicity was the conversion mechanism. A buyer arrived, saw the current tier, saw the remaining count ticking down, and made one decision. No comparison tables, no package selector, no "which tier is right for me." The visible scarcity drove urgency. 1,000 licenses gone in under 5 hours.

Node owners needed to understand what their investment was doing. License status, network contribution, reward tracking. Most node sale projects answer these questions with a Discord FAQ. I designed an admin panel that answered them in the interface. License management: visible, clear, actionable. The panel was designed as a product, not an admin screen.

Two separate design systems. Marketing design system: components and tokens for the website, optimized for visual impact and brand expression. Product design system: components and patterns for the AI Chat and admin panel, optimized for usability and extensibility, built to support features that did not exist yet. Shared foundations where appropriate (color tokens, spacing scale), divergent component libraries where the contexts demanded it.