SoundHound

SoundHound's Houndify platform put conversational AI inside products people already lived with — cars, televisions, payment systems. My job was to lead the design of experiences where the interface, most of the time, was nothing but a voice.

The platform served two audiences at once: consumers talking to their dashboards and TVs, and the Fortune 500 companies — Daimler‑Mercedes, Hyundai, Mastercard, Vizio — embedding Houndify into their own products. Every client arrived with a different context, a different brand, and a different idea of what "talking to a machine" should feel like. The design organization had to serve all of them without redesigning the world from scratch each time.

Feedback, error, and trust — without a visual channel

Screen design has a century of borrowed convention behind it: affordances, hierarchy, progressive disclosure. Voice has almost none. A user speaking to a dashboard cannot see system status, cannot scan for options, and cannot undo with a gesture. Three problems dominated everything we did:

Feedback. How does a system show it heard you, understood you, and is working — in the half-second before it answers? We designed the audible and (where a screen existed) minimal visual grammar for listening, thinking, and responding states.

Error recovery. Misrecognition is not an edge case in voice; it is the weather. Flows were designed around graceful re-prompting, partial confirmation, and never making the user repeat the whole utterance.

Trust. A voice interface asks people to act without visual confirmation — hardest of all in payments. Confirmation moments had to be explicit enough to trust, brief enough to bear.

Three arenas, three different physics

Each client category was a genuinely different interaction paradigm — not a reskin. The same platform had to feel native to a moving car, a shared living room, and a financial transaction.

One design system for every voice

Serving four Fortune 500 clients with a seven-person team only works if the design decisions compound. We built the design system used across all voice-interface implementations: the shared interaction grammar (wake, feedback, confirmation, recovery), the state vocabulary for listening and thinking, and the visual kit for the moments a screen did exist — flexible enough to wear each client's brand, strict enough to keep the conversational core consistent.

The system was also how design scaled its authority inside the organization: engineering and client teams could build from documented patterns instead of re-negotiating first principles per deal.

Player-coach, by design

I led a team of seven designers and researchers as a working director — reviewing flows, drawing the hard ones myself, and putting research directly in the loop with client engineering teams. Voice UI in 2019 had no settled playbook; the team's job was to write one while shipping against automotive and payments deadlines.

What shipped

• B2B voice implementations shipped with major automotive and payments partners, including Daimler‑Mercedes, Hyundai, Mastercard, and Vizio.
• A voice-first design system adopted across all Houndify client implementations.
• Built and led the 7-person product design and research team through the platform's key client deliveries.
• An interaction grammar for feedback, error recovery, and trust that predates — and anticipated — today's conversational-AI patterns.

Five years later, the industry caught up to the problem: every product is now learning to converse. Designing for probabilistic systems — where the interface can mishear, misunderstand, and must recover gracefully — turned out to be the rehearsal for designing with AI.