AI Evaluation
Motley
AI tool choices were guesswork — no monitoring, no evaluation, no shared numbers to compare one model or service against another.
AI Engineering · Evaluation · Tool Selection
I design and ship AI-native products from ambiguous idea to working software. My work spans product strategy, UX/UI, coded prototypes, AI workflows, and production interfaces across SaaS, adtech, dev tooling, and agentic systems. I'm strongest where design, engineering, and product judgment overlap: turning unclear technical possibilities into polished, usable products.
Make it clear, make it work, make it matter.
Recent work includes AI campaign tools, agent dashboards, MCP interfaces, LLM evaluation systems, and production workflow automation.
Sep 10 — Sep 17
Joaquin Perez — Observed Patterns — Real-Time
Selected Work
Blockboard is a streaming-TV advertising platform. Over 2024–2026 I worked across most of its surface. I also undertook branding, website building and maintenance, marketing, and solutions engineering responsibilities, though there are no case studies for these.

AI Product Design
BlockVantage
Audiences, publishers, dayparts, devices, budget: a streaming-TV campaign is too much for a first-time advertiser to set up. BlockVantage has the AI do it and shows what it recommended and why.
AI Product Design · Interaction Design · AI UX

Design Systems
Blockboard Design System
I built the design system as a web server to update the design of our new product.
Design Systems · Design Engineering · MCP Tooling

AI Product Engineering
BlockVantage Creative
BlockVantage sells streaming TV with no minimums and no experience needed — but every campaign needs a video the advertiser doesn't have, and we were never going to train a video model.
AI Product Engineering · API Integration · Product Definition

Payments Engineering
Finix Integration
A payments integration has no useful mockup — the design is whatever the API's state machine permits.
Payments Engineering · Integration Architecture · Vendor Ownership

GTM Engineering
BlockVantage.com
How do you turn a backend AI intelligence engine into a demo compelling enough to drive signups for the full platform?
GTM Engineering · Brand & Marketing Design · Video Production

Product Design
BlockConnect
Users couldn't deploy tracking pixels without calling support — the feature generated more tickets than anything else on the platform.
Product Design · UX Research · Interaction Design
AI Evaluation
AI tool choices were guesswork — no monitoring, no evaluation, no shared numbers to compare one model or service against another.
AI Engineering · Evaluation · Tool Selection
Agent Infrastructure
Agents act everywhere, but nothing recognizes them across sessions — and no one could verify an agent's identity claim without asking the service that issued it.
Designed & Developed · Protocol Implementation
Agent Tooling
The Claude Agent SDK is powerful but forces constant context-switching between code, CLI, and docs.
Designed & Developed · Side Project
Working Principles
Patterns from building — the trade-offs, the unknowns, and what changed along the way.
I cut breadth for speed. A narrow first version that works earns the credibility to widen scope later — and by then you know which extensions matter. The instinct to support everything is strong. I ship one thing well and let traction decide what comes next.
“Do one thing quickly. Expand with trust.”
When a product generates confusion, the answer is almost never more features. Fewer options, clearer labels, one obvious path forward. The best fix I have shipped was subtraction — removing steps, not adding them. Clarity compounds. Complexity just accumulates.
“The best fix was subtraction.”
Some of my strongest work started with no playbook — building on standards still being written, in territory with no tutorials and no one to ask. I read the spec, built the first implementation, and iterated until it worked. Unfinished ground is the best place to build if you are willing to write the reference yourself.
“Build on unfinished ground. Write the reference yourself.”
I build the full chain, not individual screens. That means knowing how components interact, where latency compounds, and how one decision ripples through everything downstream. The system is the product. The screen is just where you touch it.
“The whole chain, not individual screens.”
AI changed what one person can attempt. I take on work that used to require a team — and that shifted something deeper than my skill set. It shifted my ambition level. The feedback loop collapsed: building and thinking became the same motion. That is the flow state I am optimizing for now.
“Building and thinking — the same motion.”
Toolkit
The tools I reach for — and the territory I pay attention to.
Claude Code, Model Context Protocol (MCP), MCP-UI, Vercel AI SDK, Ollama, LangChain, Honcho, Local LLMs, A2A Protocol
TypeScript, JavaScript, Python, React, Next.js, Node.js, Vue, HTML, CSS, Tailwind CSS, Radix UI, Framer Motion, Storybook, Playwright
Docker, AWS, Cloudflare (Workers, KV, DNS), Vercel, PostgreSQL (Neon), Redis, Supabase, Drizzle ORM, Inngest, REST APIs, Git
Figma, Framer, Webflow, Design Systems, Interaction Design, After Effects
Decentralized Identity, Digital Sovereignty, Context Engineering, Composable UI, Open Source Tools, Media Theory