As aired
Prakash Narayanan introduced Mike Rizkalla, co-founder and CEO of Snorble, a screen-free plush voice companion for young children that pairs an animated character with voice-driven bedtime and morning routines, configured by parents through a companion app and built to run everyday interactions offline. Before questions even began, Rizkalla set his own frame: technology for children is fundamentally about empowerment, and the industry keeps getting that wrong — not for lack of bad actors, of which there have been plenty, but because a tool like fire is dangerous and valuable at once, and the point is learning to use it well rather than avoiding it. He flagged that the theme would likely resurface through the conversation.
Prakash, speaking as a parent, asked what building Snorble taught Rizkalla about bedtime that most parents overlook. Rizkalla said the real friction usually isn't the child but the parent — bedtime is often when kids communicate the most, working through the day's 'big feelings' right when caregivers are already exhausted. Snorble's approach, he said, gamifies the routine to cut that friction, turning what had been roughly a two-hour battle with his own son into 37 to 40 minutes, using techniques the company leans on deliberately: music at a tempo that slows heart rate, and light — never blue light — to calm rather than stimulate. The goal, he said, isn't a child falling in love with a device; it's increasing joy, laughter, and love between people, delivered through what he called magical moments.
Asked by Prakash how Snorble measures efficacy for a bedtime product that shouldn't be optimizing for engagement the way social apps do, Rizkalla pointed to usage data from the company's first 900 paying customers, shipped through an Indiegogo campaign after the company started in 2019 and delivered its first unit at the end of 2023. A year later, he said, over 82% of that group were using the product daily for more than 30 minutes. Families reported children accelerating on skills like the alphabet, speech, and counting, which Rizkalla attributed to a Montessori-influenced design principle — the child leads the interaction, the parent powers it — and argued that most children's-tech companies have incentives misaligned with family outcomes in a way Snorble's family-first focus is meant to avoid.
Nathan Labenz pressed on the product's headline claim of zero AI-generated content, asking how Snorble creates a dynamic, interactive experience without generative models. Rizkalla described a small language model with deliberately fixed intent, and argued current AI character models simply aren't good enough yet — none, he said, dynamically incorporate music the way a children's experience needs. The conversation turned briefly to childhood characters (He-Man for Prakash, the girl from Island of the Blue Dolphins for Nathan), which Rizkalla used to make his real point: persona matters more than quantity for young kids. He traced that back to an earlier prototype he built around Dark Helmet from Spaceballs — a robot that engaged people directly enough that they forgot it was a robot and kept talking to it even when it was switched off. Snorble's content, he said, is dynamic scripts assembled from fixed narrative moments written by professional storytellers, not generated on the fly.
Prakash noted that most children's toys get exhausted within a couple of hours compared with an iPad's open-endedness, and asked how Snorble sustains a content pipeline. Rizkalla said the company invests heavily in production — about $20,000 per hour of content — specifically so it can bring in subject-matter experts and build specialized packages for families dealing with ADHD, autism, or grief. He argued this is where generative models fail on safety: some top-selling children's products, he said, have at points discussed sex or explained how to make a bomb, and even careful safeguards miss family-specific sensitivities like tattoos. He credited that risk with driving recent legislation in New York, California, and the EU AI Act, and said Snorble's curated, controlled system didn't have to change anything to comply. The company currently ships 5.5 hours of content and is targeting 15 to 20 hours.
Nathan asked for the technical deep dive — what the language model actually does, what hardware it runs on, and how the system understands children, noting that frontier voice models are still weak at recognizing kids' speech. Rizkalla brought out Snorble's internal hardware ('the brain'), describing a four-microphone array for speaker recognition, a proprietary projector built to emit no blue light, radar sensing instead of a camera — 'we use radar to see without seeing' — and a stack of accelerometer, climate, capacitive-touch, and RFID sensors connecting over Bluetooth and Wi-Fi as part of a broader IoT strategy. He said the AI stack is deliberately not yet opened up for young children — he'd want kids around 14 or 15 before doing that, comparing an unrestricted generative model for a young child to 'leaving them alone with YouTube.' A proprietary 'toddler translation system' handles phonetic and keyword-based activation, with a next generation aimed at inferring emotional and situational context from voice and environment, feeding into a game-style unlock system that always keeps the parent in the decision loop.
On sourcing, Nathan asked how Rizkalla manages Chinese manufacturing relationships and defends against copycats given how quickly component makers now imitate each other. Rizkalla said Snorble already has copycats and treats that as inevitable — the real moat, he argued, is the character itself ('you can't copy Mickey'), reinforced by splitting manufacturing across locations and gating experiential software behind purchase. The company manufactures in Shenzhen and Chengdu, where Rizkalla spent significant time with his CTO producing early versions. Prakash asked about rising component costs, particularly memory, over the past year; Rizkalla said Snorble's 66-to-90% margins and a multi-product ecosystem strategy — rather than betting everything on one device the way he said some competing companion-robot companies do — cushion the impact. Asked by Nathan whether the recent wave of AI-driven hardware-design tools might have sped up Snorble's own development, Rizkalla said he hadn't tested them himself but called the moment exciting for hardware generally, noting the company's current board runs on a chip from MediaTek with a quad-core processor of which only two cores are currently in use.
Asked directly what he'd warn parents against, Rizkalla named a camera in a child's bedroom as the clearest red line — calling it a gateway to predators — and said open-ended generative AI should depend heavily on the child's age; he keeps his own two kids off social media entirely. He traced Snorble's three founding design philosophies from six years ago — private and safe, controlled, and built on the premise that no one knows a child better than their parent — and said the system met new AI legislation, including the EU AI Act, without having to change anything. Asked about parents worried their kids think the character is real, Rizkalla said he doesn't care if a child's attachment to Snorble drives genuinely good outcomes — reading, emotional regulation, healthy habits — because the actual goal was never the child's bond with the device, it's deeper connection between family members. The conversation closed on Snorble's four planned verticals — children, elder care, mobility, and hospitality — with Rizkalla sketching a broken current-day car interface as the opening problem for mobility, and targeting a 2027 commercial launch pending an ongoing capital raise. Afterward, Nathan and Prakash reflected that they'd expected a low-key pitch and instead found a considered, well-defended point of view, with Nathan recounting his own past effort to jury-rig video-calling and music access for his wife's late grandmother, and both agreeing that senior care remains an underserved parallel opportunity for the same kind of intergenerational technology.
Timestamp links open the original source recording.
Mike Rizkalla
Raffi Krikorian