Sparkli Rolls Out Multimodal AI for Child-Led Learning

Zurich-based Sparkli, founded by former Google and YouTube veterans Lax Poojary, Mynseok Kang, and Lucie Marchand, has exited stealth with $5 million in pre-seed funding. The company is developing a multimodal, AI-native learning engine aimed at children aged 5 to 12, designed to transform their questions into interactive, multidisciplinary learning journeys. The platform moves beyond static text-based responses, a limitation CEO Lax Poojary observed when using tools like ChatGPT and Gemini with his own son. Instead of a "wall of text," Sparkli generates immersive "expeditions" that blend visuals, voice interaction, and playable simulations to foster skills like problem-solving and creativity. For example, a question about building a city on Mars could generate an interactive experience involving physics simulations and infrastructure design. Underpinning Sparkli's adaptive capabilities is a system that builds an evolving interest and knowledge graph for each child. This allows for the personalization of content over time, a core challenge in edtech that can be addressed with knowledge tracing models. Such models can infer a learner's knowledge state to predict future performance and tailor the learning path accordingly. To dynamically optimize these learning paths, reinforcement learning (RL) offers a powerful approach for sequencing educational activities. An RL agent could learn to recommend the next step in a learning "expedition" to maximize engagement and learning outcomes, adapting in real-time to a child's interactions. This is often framed as a multi-armed bandit problem, where the system explores different content recommendations to see which yields the best response. A key modality for Sparkli is voice, which presents unique challenges with young learners due to variations in vocal tract length, pronunciation, and vocabulary. Overcoming high error rates in automatic speech recognition (ASR) for children is critical for creating effective voice-driven tutoring systems and adaptive literacy assessments. Advances in ASR now include custom models tailored to the acoustic properties of children's voices. With a user base of young children, AI safety is a paramount concern. This involves not only content filtering and parental controls but also designing for age-appropriate interactions and protecting children's data. The platform's design must prioritize trust and safety to prevent exposure to misinformation or harmful content. The user experience for children requires a different approach than for adults, emphasizing simple, intuitive interfaces with large, tappable elements and immediate, positive feedback. For multimodal learning, balancing sensory inputs like sound and animation is crucial to maintain engagement without overwhelming the user. Sparkli is currently testing its platform in a pilot program with a large private school group, giving them access to over 100 schools and 100,000 students. A private beta launch was planned for January 2026, with a broader consumer rollout scheduled for June 2026.