Profile
The infrastructure architect who keeps arriving early
There is a specific kind of founder who doesn't discover a trend but instead has been quietly waiting for the world to catch up to them. Kwindla Hultman Kramer is that kind. When most developers were still copy-pasting YouTube embeds, he was building real-time video infrastructure on Nokia 6600 handsets. When the AI boom made voice agents fashionable in 2023, he had already spent seven years running the pipes beneath them.
Today he is co-founder and CEO of Daily.co, a developer platform that handles real-time voice, video, and AI communication infrastructure for companies that would rather not rebuild WebRTC from scratch. That is most companies. Daily's WebRTC network sits between users on devices and the servers generating audio and video - a layer of plumbing so fundamental it barely shows up in product screenshots but shows up constantly in latency numbers and uptime logs.
"Don't start a company because you like building things - start one because you want to build a company that's good at building things."- Kwindla Kramer, on the difference that matters
Beyond Daily sits Pipecat, the open-source framework Kwindla created for building voice AI agents. It is, by most measures, the most widely used framework of its kind. Pipecat handles the parts of a voice conversation that are deceptively hard: interruption handling, turn-taking, multi-turn context, modular model swapping. The details that make the difference between a voice bot that feels like a tool and one that feels like a conversation.
Origin Story
Three degrees of infrastructure
Kwindla's career reads like a series of bets on what communication technology would look like before communication technology looked like that. He came out of MIT's Media Lab with an MS in 1999, after an AB from Harvard - but what shaped him wasn't the credential, it was the environment. Media Lab professor Mitchel Resnick's work on tools for learning and collaboration pointed him toward systems designed to connect people rather than process data.
His first major bet was AllAfrica.com, which he co-founded as CTO the same year he left Cambridge. AllAfrica became one of the web's largest content aggregators - thousands of stories per day, millions of monthly visitors - at a time when most of Africa's press was invisible to the global internet. The infrastructure challenge was real: custom editorial tools, content management systems built for scale, a web backend that had to be reliable across unreliable connections. He remains on the company's board today, 25 years later.
In 2005, he moved across the country to Los Angeles to help build Oblong Industries. As CEO, he scaled the company from 2 to 80 employees over eight years. Oblong's work on the g-speak Spatial Operating Environment - gestural, spatial computing interfaces before that phrase became a buzzword - positioned the company as a serious player in high-end enterprise technology. This was pre-iPhone, pre-touch, pre-Minority Report being treated as a product roadmap. Oblong was doing it for real.
"Traction is the best way to provide evidence of future success. Month-over-month revenue growth at non-trivial scale is the best metric to discuss."- Kwindla Kramer, Y Combinator advice
Daily.co started in 2016 with a Y Combinator batch (W16) and a clear thesis: every application would eventually need real-time video embedded directly into it. Co-founded with Nina Kuruvilla, the company went from idea to product to $6M ARR and over 1,000 customers by 2024. The Series B in November 2021 - $40M, led by Lachy Groom - validated what the growth numbers were already saying. Half of Daily's users are outside the US. On day one, they got international orders. Kwindla had been thinking about global infrastructure from the start.
The Pipecat Story
Open-sourcing the hard part of voice AI
When large language models got fast enough to have a real conversation with, the voice AI problem became visible to everyone at once. But the people who had been watching it knew exactly how much was still missing. The models were there. The infrastructure wasn't.
Kwindla's answer was Pipecat - a full-stack framework for building voice and real-time multimodal AI agents. Not a wrapper around a single model API, but an orchestration layer that handles the actual complexity: how do you manage interruptions when both the user and the bot want to speak? How do you handle turn-taking that feels natural rather than robotic? How do you swap between different speech-to-text, LLM, and text-to-speech models without rebuilding the whole pipeline? How do you get end-to-end latency under 500ms so the conversation doesn't feel like a phone call on a bad connection?
These are not glamorous problems. They don't generate Medium thinkpieces. They are exactly the kind of problems that, when solved, make everything else possible. Pipecat solved them and released the solution to the open-source community. The adoption was immediate.
The framework takes a modular, multi-model approach - the kind of system that production deployments actually need, where different tasks call for different specialized models and you can't afford the latency of routing everything through one monolithic API. Kwindla's position on this is consistent: favor practical over elegant, production-ready over demo-ready. His talk at the AI Engineer World's Fair on building the world's fastest voice bot was not a pitch for a product. It was a tour of the actual engineering decisions that shave hundreds of milliseconds off response time.
Philosophy & Craft
Building companies that build things
The through-line in Kwindla's career is a particular kind of ambition - not to be known for a product, but to be known for an organization that reliably ships hard things well. His writing on the Y Combinator blog about building startup executive teams is unusually unsentimental about this. "Executives must fit the company, stage, and market." Not fit in general. Fit now. For this moment. The right great person for a Series A is a different person than the right great person for a Series B.
He's equally clear about what investor conversations require: traction first, everything else second. Month-over-month growth at non-trivial scale. Knowing your basic accounting. Understanding that a letter of intent is not a contract. The practical, specific knowledge that separates founders who've done it from founders who are explaining it.
On technology: avoid both excessive optimism and undue pessimism about what's emerging. The Nokia 6600 video call wasn't the future - but the direction was right. The LLM-powered voice assistant wasn't ready in 2020 - but the direction was right. Kwindla has been consistently right about the direction long before the timing worked out. That's the harder skill.
He mentors startups through Mucker Capital in Los Angeles and remains active in the developer community through a GitHub that includes projects ranging from local voice agent implementations to a GPT-powered open-source space game. He is not a founder who gets too big to write code. The hands stay in.