I’m David Paterni, a product designer for complex SaaS and enterprise workflows.
I turn messy product logic, dense data, and hard constraints into clear experiences people trust.
Case Studies
Shipped and concept product work across consumer and enterprise. Each one carries the problem, the constraints, the tradeoffs, and the interface decisions, not just the final screens.
Next.js + TypeScript + Tailwind, on Vercel
Criterion: The AI interview coach that shows its work.
Watchtower, live agent supervision
Supervise an autonomous voice agent while it is on the call. An act window catches a consequential action a beat before it fires, scope sits beside the live transcript, and some gates, like patient consent, stay closed even under a human takeover.
Conformly, EAA compliance agent
Accessibility, delivered as evidence. Deterministic detection mapped to EN 301 549, fixes shipped as real pull requests, and a hash-chained, tamper-evident audit trail. It never says you're compliant; honest status is the product.
Notification Sieve
A calm, context-aware filter that intercepts the noise, escalates only what matters, and lets you catch up on your terms.
Turning overlooked behavior into a new product opportunity
How identifying a real world signing gap led to a high impact proof of concept for KeySign.
1QR
Scan once. See the real total. Pay before anything moves.
Programmatic SEO that earns its long tail
How Much Would I Have If: a what-if investment calculator, ~1,070 static pages built to rank honestly.
Vantage, time to insight
Observability. Turning a sea of fragmented signals into one line to the cause: hand-built accessible topology and flame-graph visualization, an AI read-lane, and a deliberate handoff to Oversee for execution.
A review score you can't buy
DreadRobot: skeptical AI tool verdicts. A five axis score, hard caps for tools that lie, integrity enforced in the build.
Range, iOS rehab form coach
A doubt removal machine for solo knee rehab. On device pose estimation corrects form in real time without making a vulnerable person feel watched, and shows range of motion climbing week over week. Three live interactive moments inside.
Readout, data and dashboard design
Data visualization. The judgment behind a good dashboard: the right chart for the question, honest scales, drill-down on demand, and accessibility built in, not bolted on.
The Interaction Layer, a masterclass in how software feels
Seven principles of interaction design: feedback, perceived performance, affordances, reversibility, disclosure, direct manipulation, and input, each grounded in HCI law and proven with a live, hands-on demo.
Transforming customer context into actionable clarity
A case study on building a unified context feed that helps agents resolve conversations faster, powered by replay intelligence.
Designing clarity into large-scale personalization systems
A design-led approach to explaining model behavior, simulating change, and enabling confident decisions.
Designing real estate moments people can trust
A deeper look at how I transformed familiar home-search patterns into a high-trust experience, centered around Sneak Peek listings.AI Product Design Lab
Focused probes into the hard parts of agentic AI. Setting intent before a model generates, verifying what it claims, supervising autonomy that runs on its own, and making memory something you can see and control. Each one is a working concept you can click.
Steer, intent before generation
A blank prompt invites a confident guess. Steer negotiates the brief first, surfacing the audience, goal, and tone the model would otherwise fill in silently, so intent is set before generation, not corrected after you ship.Ground, verify what AI claims
A fluent paragraph reads as one confident thing, but it is really a mix of well sourced, stale, contested, and invented. Ground shows what each sentence actually rests on, so fluency stops getting mistaken for accuracy.Oversee, safe agent autonomy
The move from chatbots to agents is the move from systems that talk to systems that act, the ones that send the money and change the records. Oversee is a supervision layer for autonomy that runs for minutes across many steps, past the point where confirmation prompts can keep up.Recall, legible AI memory
Personalization is supposed to help, but it accumulates in the dark. Recall makes AI memory legible: see what an assistant has stored, scope what it can use, and revoke what it keeps.Playbooks
Reusable operating standards drawn from how the work actually runs. Running Figma with an agent in the file, writing interface copy as one voice, and taking a vague request to a buildable plan.
AI Operating System
How I work with AI across the whole practice. The seven ways AI drifts toward plausible, the rules that hold intent fixed, the context I brief it with, and the verification gate that the machine never gets past.Discovery to Scope
How I take a vague product request to a plan a team can build. Jobs to be Done to find the real job, the Double Diamond to widen then narrow, and a scope document that names the problem, the bet, and what is out.Figma Operating System
An operating standard for running Figma with an agent in the file. Naming, status, tokens, versioning, handoff, and the context model that makes generated work match your system. Written for power users.Interface Content System
The standard I apply to interface copy on every screen. Customer perspective over system perspective, active controls, positive framing, and accessible mechanics. Every rule shown as a before and an after.Design Systems
Operable design systems, not slide decks. Real tokens, components, blocks, and full app screens, built to hold up under real product states, accessibility, and the messy edges of the work.

Kinetic Design System
A consumer banking design system built around motion and clarity, with real tokens, components, and screens where money that moves feels like it.
Kernel Design System
Design system. Live, fluid, in light and dark, with real tokens, components, blocks, and screens.