AI Works & Pet Projects

Situation Quadcode was building one AI platform for coding, design and content creation, and the MVP needed a foundation for it: code and media agents in one place.

Task Design the entire product interface from scratch, build the visual identity, and make the agent workflows readable end to end.

Action I ran a UX audit and built the MVP brand. I designed the product interface, the design system behind it, and the workflows that put code and media agents in one environment.

Result Delivered the product design, core design system and branding for the MVP release: one environment where developers and creators run code and media agents. The platform is public and running — quadcode.ai.

• Git-backed Truth: The entire project state (roles, process, history, budget) is in the repo. Clone the repo, and the whole project comes with it.
• Granular Cost Tracking: Costs are tracked per task, role, and release, exposing what was spent on work that produced nothing.
• Resilient Execution: A run that dies resumes from its state inventory instead of restarting. Applied work is never redone.
• Earned Autonomy: Trust and autonomy move based on evidence in run records. Agents never loosen their own gates.
• Honest Enforcement: every gate names what holds it — validator, hook, git host, or prose-only, listed by name. The rules that mattered were measured from 0/5 as prose to hard refusals as hooks.
• One Repo, Every Runtime: installs as a plugin into Claude Code, Codex, Gemini CLI and more — the project travels with its repository, not with a tool.

• Computed Output: Low-level building blocks rather than a fixed high-level scheme — you set the model, the tokens follow.
• Multi-Agent AI: A built-in AI chat with dedicated personas handles audits, proposals, and hands-on theme editing.
• Full Token Coverage: Computes every token in the model across light/dark modes, conditions and interaction states, including inverse tokens.
• Built-in Accessibility: Contrast ratios are part of the model, and vision filters simulate how the system appears to users with color vision deficiencies.

Situation I joined a Stealth AI Startup as Head of Design in a small, cross-functional team to build custom AI companion products.

Task The objective was to design both the B2C product (end-user companion) and the B2B platform (white-label system) with a design process built for speed.

Action I built all design processes from scratch, embedding AI into the workflow. This included content generation via custom node-based pipelines and Figma Weave, and asset generation via custom bash scripts that saved ~3 months of manual work. All competitor analysis, copywriting, mood boarding, persona/scenario creation, and hypothesis validation were AI-assisted. I also contributed to frontend development for the MVP.

Result Shipped the MVP with direct frontend contributions. The B2C companion reached 1M+ total traffic, 100K+ MAU and 40K+ DAU; the B2B white-label platform shipped to partners. Automated asset generation — ~3 months of manual design and engineering effort saved.

• Closed-Loop Conveyor: A dedicated conductor runs every stage — discovery, design, parallel review gates, shipping, measuring — instead of stopping at a merge.
• Cost & Governance: Every release writes an itemised cost ledger — tokens, time and list-price dollars, per agent and per feature — from the platform's own counters, so "what did this cost" has an answer instead of a feeling. A declared budget shapes what gets proposed, not just what gets refused — and spending is one of four kinds that always route to you for a yes, however long the track record.
• Long-term Resilience: Session-limit recovery, health checks across the whole conveyor, and git-backed plugin & skill upgrades.
• Honest by Construction: Every rule states what actually enforces it — a request, a validator, branch protection, the platform, or prose-only, which means nothing does — and the prose-only rules are listed by name. Behaviour is measured against stratified scenarios, each release publishes its run record including what failed, and a coverage map says which scenarios are unmeasured rather than implying all of them pass.
• Autonomous Staffing: A built-in role-builder researches crafts, drafts skills, and hires specialized agents (or invites human teammates) to grow the team.