AI Lab
Working AI Products, Built to Production Standards.
I build, test, and iterate to production standards. I use AI to build AI: computer vision, behavioral agents, LLM interfaces, and governance. The best argument for an idea is a working system.
WEB APP · JEV · TYPESAFE AI
Jevit
Jevit turns invoices and receipts into structured data, with every field linked back to its source in the document. Extraction happens locally, while Jev checks the document type and verifies whether the cited evidence supports each value.
PLUGIN · CLAUDE AI · AGENTS
PingBack
PingBack turns workshop boards into live facilitation systems. It connects to Miro, reads sticky notes, and uses Claude agents to generate one critical-thinking question per idea. Participants respond from their phones without interrupting the room, and agents read those signals to ask smarter follow-up questions.
AI GOVERNANCE · FRAMEWORK
Reversion
Reversion AI is a 6-step framework and practical planning canvas experience that makes reversibility a structural requirement in AI interactions. By design, every system action can be undone, audited, or redirected by the user at the interface layer.
HCI · AI INSIGHTS
SpotDiary
SpotDiary maps remote work spots through four behavioral dimensions. Every check-in feeds a radar profile and an LLM that surfaces patterns the user never noticed about their own work habits. The spot is the input. The insight is the output.
AI AGENT · NLP · BEHAVIORAL SIGNALS
Veltra
Veltra detects typing friction in real time (words per minute, idle time, backspaces, focus switches) and triggers proactive channel switching before abandonment happens. Behavioral signals, not explicit commands, drive the handoff logic.
NLP · COMPUTER VISION · HEALTHCARE AI
RX Reader
RX Reader uses computer vision and medical language models to interpret handwritten prescriptions and surface reliable medication information to pharmacists in seconds. Unclear handwriting is the input. Clinical confidence is the output.
CODE · COMPUTER VISION · LLM
UX Wise AI
UX Wise AI combines vision models and LLMs to analyze UI screenshots and surface what code review never catches: UX flaws, accessibility gaps, behavioral drift, cultural mismatches, and privacy red flags. A reasoning partner that reads interfaces like a senior designer and ships the audit in seconds.
PROMPT ENGINEERING · GUARDRAILS
Shard Protocol
Shard Protocol applies pre-execution prompt fragmentation, constraint mapping, and logic path validation to enforce safe, verifiable AI behavior. The protocol enforces what most AI interfaces assume: that the model correctly understood the intent before acting.
AI AGENT · APPLESCRIPT · API · HARDWARE
Clippy AI
Clippy AI connects ChatGPT to a physical keyboard via AppleScript, triggered by a single keystroke, in a 3D-printed prototype. One of computing's most mocked interfaces, rebuilt with a real brain. A provocation about what ambient AI assistance feels like when it is always one gesture away.
GENERATIVE AI · PROMPT ENGINEERING
Re:Branded by AI Prompts
Re:Branded stress-tests AI as a creative partner in brand work. Built from over 200 curated prompts that explore a single identity across its full aesthetic range. The output proves what structured prompt engineering can do when applied with design discipline.
HCI · CONVERSATIONAL DESIGN
SmartEats
SmartEats is a concept for the AI Pin that replaces voice-based allergen detection with visual detection. Color-coded icons identify food intolerances in real time, making lunchtime safer and more independent in noisy public spaces. A design exploration proving that inclusive AI interfaces require intentional constraints.
Curious about some of my side design projects?
Feel free to hit me up. I'm looking forward to hearing from you.
hello@marcosrezende.com
Marcos Rezende · Portfolio · UX / Product Design / AI
© 2026 · Privacy policy