The Infrastructure Bet That Keeps Paying Off
Nina Kuruvilla co-founded Daily.co in 2016 with a fairly specific ambition: make it easy for any developer to add real-time video to any application. Not a consumer product. Not a Zoom competitor. Pure infrastructure, offered as an API. The sort of company that nobody writes magazine profiles about until, suddenly, the whole world depends on it.
The pandemic made Daily visible. Between 2019 and 2020, the company logged 30x revenue growth and a 20x increase in paying customers - not because they pivoted, but because the infrastructure they'd been quietly building for four years turned out to be exactly what the world needed. Telehealth platforms, remote-work tools, developer teams that had been sleeping on real-time video - they all showed up at once.
But the more interesting part of the Daily story isn't the pandemic spike. It's what Nina recognized next. The same WebRTC infrastructure Daily had built for humans talking to humans worked, with striking fidelity, for humans talking to AI agents. Low latency. Reliable audio pipelines. Global mesh network. All of it.
"This infrastructure we had built for human-to-human communication," she said, "was the perfect infrastructure for enabling humans to talk to LLMs."
That observation became Pipecat - an open-source Python framework for building real-time voice and multimodal conversational AI agents. Daily open-sourced it. Intentionally vendor-neutral - meaning developers can run Pipecat on Daily's infrastructure, on competitors' infrastructure, or on no cloud at all. The board initially questioned that choice. Nina pushed forward. NVIDIA uses it. AWS uses it. Thousands of startups use it to build voice agents for healthcare, customer service, telehealth, and interactive AI systems.
"You can use Pipecat with Daily's network stack, of course. But you can also use Pipecat with completely Open Source tooling, with your own custom infrastructure, and with our competitors."- Nina Kuruvilla, Co-founder, Daily.co
The strategic logic is classic infrastructure thinking: own the layer the ecosystem depends on, make the ecosystem as large as possible, and let adoption do the work. It's why AWS open-sources tools it doesn't need to, why Stripe publishes libraries for languages it doesn't profit from directly. Nina understood that the developer community around Pipecat would matter more than any exclusive advantage Daily could extract from keeping it proprietary.
What Daily and Nina have built is not just a product - it's a position. In the stack between the voice AI model and the human speaking into a phone, Daily's infrastructure is the part that makes it work in real time, at scale, with enterprise-grade compliance. HIPAA. SOC 2 Type 1. End-to-end encryption. The stuff that healthcare companies and financial services firms require before they'll run anything voice-adjacent on an external platform.
From Internal Tooling to NVIDIA's Stack
Pipecat started as internal Daily tooling - the kind of glue code that accumulates in any infrastructure company as developers solve the same problems over and over. Somebody needs to chain together a speech-to-text model, a language model, and a text-to-speech engine in a way that doesn't introduce 800ms of latency. Someone else needs turn detection. Someone needs to handle interruptions mid-sentence.
Nina recognized that every team building voice AI was writing the same code. Pipecat abstracts that. It handles the orchestration layer - the timing, the streaming, the model switching, the session management - so developers can focus on what their agent actually does rather than how it handles audio frames.
The open-source hackathon for Pipecat grew from 40 registrations to 564 in one cycle. That's not just a growth metric - that's evidence of a framework hitting product-market fit with developers in real time.
Nina has also been revising the prevailing wisdom on model architecture: the "one big model" approach is giving way to multi-model pipelines where specialized models handle specific tasks in concert. Pipecat is built for that. It was always built for that - the framework assumes you'll be chaining models, not relying on one.
"What we see as we work with teams is an emphasis on collaboration and flexibility. There is so much creative capital and human potential across different ecosystems."- Nina Kuruvilla
Eight Years Building Real-Time Infrastructure
What She Built
-
★Co-founded Daily.co, raising over $66M in total funding from Y Combinator, Lachy Groom, and Renegade Partners
-
📈Guided Daily through 30x revenue growth and 20x paying customer growth during the 2020 pandemic
-
🔗Championed open-sourcing Pipecat, now adopted by NVIDIA, AWS, and thousands of enterprises and startups worldwide
-
🔨Built infrastructure supporting 15,000-person video calls - 300x the 50-person industry average
-
🌍Established enterprise-grade compliance: HIPAA, SOC 2 Type 1, end-to-end encryption for real-time audio/video
-
🎉Grew Pipecat developer community from 40 to 564 hackathon registrations in a single cycle
-
🧹Led Daily's strategic pivot from human-to-human video to real-time AI voice agent infrastructure
Building a Company With Your Spouse, Remotely, in Real Time
Daily.co is a remote-first company. That's not a work-from-home policy - it's an architectural decision. The company was distributed before the pandemic made that fashionable, and it's shaped everything from how they communicate to what they believe about the future of collaboration.
Nina runs the company with her co-founder and spouse, Kwindla Hultman Kramer. On the question of how that works, she keeps it practical: "We try to be intentional about work-life balance." Not a manifesto. Not a Ted Talk. A sentence that acknowledges the difficulty and the choice.
"We want to be a place where everyone feels they have expertise they are uniquely able to teach their colleagues about."
Nina Kuruvilla on Daily's culture
She has also made a point of saying that writing matters in a video company - perhaps counterintuitively. VentureBeat ran a piece on this: Nina's argument that a return to writing is vital for Daily, that the discipline of putting ideas into sentences is a competitive advantage even when your product is video. The kind of contrarian take that usually turns out to be correct.
On social media, she's been deliberate - prioritizing X and LinkedIn as venues for substantive conversations about LLM applications and voice AI infrastructure. Not performative presence. Actual signal in a space full of noise.
She now follows industry debates about whether multi-model architectures will outpace single-model approaches. Her read: the "one big model" era is transitional, not terminal. Pipecat was designed for a world where you chain models together. That world is arriving faster than most people expected.