Roark Emerges as the QA Layer Voice AI Has Been Missing
Roark, a Y Combinator Winter 2025 company, is betting that the difference between a novelty and a trustworthy product comes down to rigorous, automated testing — and it is building the platform to make that possible. As enterprises race to deploy voice agents for customer support, sales, and operations, a quieter problem has surfaced behind the hype: reliability. Voice AI sounds magical in a demo and unpredictable in production.
Founded in 2025 and based in San Francisco, Roark gives voice AI teams the monitoring, evaluation, and simulation tools they have been cobbling together with spreadsheets, ad hoc scripts, and weekend hackery. In its first six months of operation, the company has already processed more than 10 million minutes of calls, powering quality assurance for teams across the voice AI ecosystem, including a growing roster of fellow YC startups. The platform offers more than 40 built-in metrics — measuring latency, instruction-following, sentiment, and conversational integrity, among others — alongside customizable dashboards, alerts, and the ability for teams to define their own evaluation criteria. It can automatically identify up to fifteen distinct speakers in a single call, a capability that becomes essential when monitoring complex multi-party conversations.
Where Roark really separates itself is in its simulation engine. Teams can define end-to-end phone or WebSocket simulations for both inbound and outbound agents, configuring personas with different accents, languages, speech patterns, and behavior profiles. Tests are written as conversation graphs, letting engineers branch into edge cases and difficult variants without rewriting flows from scratch. Crucially, every failed call in production can be converted into a repeatable test, closing the loop between observation and improvement so that each failure quietly hardens the agent against future regressions.
The company is led by founders with deep operational chops. CEO James Zammit spent more than a decade in infrastructure and AI engineering, most recently as a senior engineer at AngelList where he helped build core portfolio infrastructure as the firm scaled from $10 billion to $124 billion in assets under management and led the development of Relay, an AI-powered portfolio manager processing thousands of financial documents each month. Co-founder Daniel Gauci brings a similarly strong track record, having scaled Akiflow (YC S20) to $1.5 million in ARR and over 10,000 customers, with seven prior years at Casumo where he led mobile development for an app used by millions of players. The pair launched Roark on Product Hunt and have been steadily expanding the platform’s footprint ever since.